Summer Research Paper. Investigating the Role of Financial Analysts in the Internet Bubble

Transcription

1 Summer Research Paper Invesigaing he Role of Financial Analyss in he Inerne Bubble Conduced by: Yao Tian Supervised by: Pa O Brien Summer, I hank Pa O Brien for her insigh and research guidance and Sean Speers for his echnical help. 1

2 Table of Conens Absrac Inroducion Lieraure Review and Hypohesis Developmen Analyss Forecas/Recommendaion Opimism Inerne Sock Overpricing Hypohesis Developmen Research Design Sample Selecion Daa Collecion The Logisic Regression Tess for H The Logisic Model The Yearly Analysis The OLS analysis Tes for H Conclusion References

3 Invesigaing he Role of Financial Analyss in he Inerne Bubble Absrac This paper empirically invesigaes he role of financial analyss in he Inerne Bubble. I addresses he following wo research quesions: (1) were financial analyss relaively more opimisic abou Inerne socks han oher socks during he Inerne Bubble period and (2) did heir relaive opimism for he Inerne socks inflae he prices of hese socks o creae he Inerne Bubble. Using a Logisic Regression of recommendaions on indusry secors, I invesigae analyss relaive opimism for he Inerne socks while conrolling for heir over-opimism for all socks in general and heir relaive opimism for IPO socks in paricular. The resuls sugges ha during he Inerne Bubble period, financial analyss made relaively more opimisic recommendaions for he Inerne socks han for oher socks; his relaive opimism is specific o Inerne sock and does no apply o oher socks in he high-ech secor. To furher invesigae he impac of his relaive opimism, I use an OLS regression o examine wheher analyss recommendaions for he Inerne socks drove he prices of hese socks. The resul suggess ha analyss recommendaions have no significan impac on he socks fuure prices. I hus conclude ha alhough financial analyss were relaively more opimisic for he Inerne socks, heir relaive opimism did no inflae he prices of hese socks o creae he Inerne Bubble. 3

4 1. Inroducion Inerne socks firs wen public in he early 1990s. Wih he success of he early Inerne iniial public offerings (IPOs), such as Yahoo (1996), Amazon (1997), and ebay (1998), an increasing number of Inerne firms joined he marke in he lae 1990s. In 1999 and 2000 a oal of 338 Inerne companies wen public, more han doubling he number of exising Inerne IPOs. Even more asounding han he growh of Inerne IPOs during hose 2 years was he dramaic rise in share values for Inerne socks. During he wo-year period , reurns on an Inerne index (INTDEX) compiled by Pegasus Research Inernaional exceeded 125% while he NASDAQ and S&P 500 indices grew a 85% and 19.5% respecively (Liu and Song, 2001). However, he remarkable rise of Inerne socks was eclipsed by heir monumenal failure. Ending he year 2000 a and , he Inerne and NASDAQ indices dropped by more han 19% and 51% respecively (Hiriam, 2000). This represens he bigges decline in sock indices since he Grea Depression in 1931 (Hiriam H., 2000). Afer he bubble burs, we were lef wih he unresolved quesion: Wha caused he Inerne Bubble? Some major financial observers placed he blame on financial analyss for misleading invesors during he Inerne Bubble. As Liu and Song (2001) quoed from he New York Times (December 31, 2000), insead of providing invesors wih he kind of analysis ha would have kep hem from marching over he cliff, analyss prodded hem forward To provide furher evidence for/agains his claim, his paper empirically invesigaes he role of financial analyss in he formaion of he Inerne Bubble. I addresses he following wo relaed research quesions: (1) Were financial analyss 4

5 relaively more opimisic for Inerne socks han for oher socks during he Inerne Bubble period? and (2) Did heir relaive opimism (if i exised) inflae he prices of Inerne socks o creae he Inerne Bubble? Prior research (i.e. Liu and Song, 2001) documened analyss over-opimism during he Inerne Bubble and concluded ha analyss should be held responsible for he formaion of he Inerne Bubble. However, here is a gap beween heir empirical resuls and heir conclusion. Pas lieraure on analyss forecas behavior suggess ha analyss end o be over-opimisic abou socks in general (e.g. Francis and Philibrick, 1993) and ha he exen of forecas over-opimism is he greaes in he firs year following a sock s IPO (e.g. Rajan and Servaes, 1997). In falling o conrol for hese alernaive causes of forecas opimism, Liu and Song do no provide convincing evidence for analyss specific opimism for Inerne socks. Furhermore, o conclude ha analyss should be held responsible for creaing he Inerne Bubble, one also has o show ha heir relaive opimism for Inerne socks (if i exised) indeed inflaed he prices of hese socks. Taking hese facors ino consideraion, his paper uses a Logisic regression o examine he exen of financial analyss relaive opimism owards Inerne socks, and i employs an Ordinary Leas Square (OLS) regression o furher invesigae he impac of his relaive opimism in he formaion of he Inerne Bubble. To provide more compelling empirical evidence, I conrol he alernaive explanaions for analyss opimism hrough he following research design. I divide he sample ino hree indusry secors: Inerne, non-ne high-ech (NNHT), and non-ech (nontech). While examining analyss relaive opimism for he 5

6 Inerne socks, I use he nontech secor as he base group o accoun for analyss overopimism for socks in general. I use only IPO socks of he same period as samples o conrol for ime-period-specific effecs and age-of-company effecs. Also, I include pas sock performance, fuure earnings performance, fuure sock performance and IPO-year as conrol variables o ensure ha any observed opimism is no confounded by hese facors. Resuls of he Logisic regression show ha analyss were indeed relaively more opimisic abou Inerne socks han abou oher socks; his relaive opimism is specific o Inerne socks and does no apply o oher socks in he high-ech secor. The OLS regression analysis suggess ha during he Inerne Bubble period analyss recommendaions had no significan impac on he fuure prices of he Inerne socks. Based on hese resuls, I conclude ha alhough financial analyss were relaively more opimisic abou Inerne socks, heir relaive opimism did no direcly cause he formaion of he Inerne Bubble. The remainder of his paper is organized as follows: Secion II reviews relaed lieraure and develops hypoheses abou analyss relaive opimism for Inerne socks during he Inerne Bubble period. Secion III describes he sample and research design. Secion IV and V presens he Logisic and OLS regression analyses, and secion VI concludes he sudy. 6

7 2. Lieraure Review and Hypohesis Developmen 2.1 Analyss Forecas/Recommendaion Opimism A large number of empirical sudies has repored evidence ha analyss forecass are opimisic (e.g. Barefield and Comiskey, 1975; O Brien, 1988). More recenly, using IBES forecass from 1983 o 1996, Lim (1998) repors saisically and economically large forecas opimism, pervasive in ha i is observed every year in every marke capializaion decile of sample firms (Kohari, 2001). Prior sudies have aemped o explain he naure and cause of his observed opimism. Darlin (1983) argues ha analyss add bias o heir rue beliefs and repor opimisically because hey fear jeopardizing poenial invesmen banking business; Dirks and Gross (1974) argue ha hey do so because hey are concerned wih losing access o managemen as a source of informaion. On he oher hand, McNichols and O Brien (1997) suggess ha analyss selecive coverage for firms ha hey perceive as having favorable fuure prospecs serves as an alernaive explanaion for he observed opimism in heir recommendaions and earnings forecass. This self-selecion model suggess ha analyss hold a relaively more opimisic view for socks hey decide o add han for socks hey decide o drop and hey are relaively more opimisic a he iniiaion of coverage for a sock and relaively more pessimisic near he end of coverage. The IPO lieraure provides an alernaive explanaion for analyss relaive opimism for new socks. A number of academic sudies have examined he relaionships among IPO underpricing, afermarke performance, and analys coverage: Bradley e al. (2003) find ha analyss almos always iniiae coverage wih a buy or srong buy recommendaion; 7

8 Dechow e al. (1997) and Rajan and Servaes (1997) sugges ha sell-side analyss longerm growh forecass are sysemaically over-opimisic around equiy offerings; going a sep furher, Lin and McNichols (1998b) documen ha over-opimism is he greaes for underwriers and Michaely and Womack (1999) argue ha underwrier analyss give IPO socks a booser sho (posiive recommendaions) wihin he firs year of heir IPO. In summary, hese sudies sugges an IPO effec - analyss repor more opimisically around he sock s iniial and seasoned equiy offerings. 2.2 Inerne Sock Overpricing The lieraure suggess ha he valuaion model for Inerne socks is immaure and ha Inerne sock pricing is irraional. Cooper e al. (1998) repor ha 95 firms ha changed heir name o include.com or Inerne beween June 1, 1998 and July 31, 1999 experienced an average abnormal reurn of 89 percen for he en days surrounding he announcemen. Along he same lines, Cornell and Liu (2000) idenify several cases in which he marke value of a paren company was less han he marke value of is Inerne subsidiary. To explain his phenomenon, Hand (2000) proposes ha he apparenly irraional pricing of Inerne socks - he negaive correlaion beween marke value and earnings - may be due o he differen valuaion models underlying Inerne sock pricing. Trueman e al. (2000) goes a sep furher, documening ha for e-ailers and poral and communiy Inerne firms, ne income is negaively relaed o marke value, bu when ne income is decomposed, marke value is posiively relaed o boh gross profi and research and developmen expendiures. Due o a lack of pas earnings hisories and well-esablished valuaion models for Inerne socks, forecasing he earnings and fuure growh poenial of hese socks is 8

9 difficul. Some financial press (e.g. The New York Times) argue ha analyss irraional forecass and recommendaions for Inerne socks caused he Inerne Bubble. Moivaed by his conjecure, Liu and Song (2001) provide empirical evidence for analyss conribuion o he Inerne Bubble. Liu and Song examine analyss forecas behavior prior o and during he Inerne Bubble. They use he burs of he Inerne Bubble, March-April 2000, as he even. Employing he Inerne IPOs as samples, hey collec and analyze analyss earnings forecas daa for he ime period wo quarers before and one quarer afer he even quarer. On he basis ha he quarerly earnings forecas errors are significanly greaer han zero in he wo quarers prior o he marke crash in April 2000 and less so during and afer he crash, hey conclude ha financial analyss were oo opimisic abou Inerne socks and should be held accounable for creaing he Inerne Bubble. However, here is a gap beween heir empirical resuls and heir conclusion. The empirical resuls sugges ha analyss were over-opimisic abou Inerne socks before he marke crashed. However, his does no flow logically o he conclusion ha hey share some blame for he formaion of he Bubble. Prior lieraure on analyss forecas behavior suggess ha analyss end o be over-opimisic abou socks in general and ha he exen of forecas over-opimism is greaes in he firs year following a sock s IPO. Thus, he forecas opimism repored by Liu and Song may be due o analyss overall opimism and/or IPO effecs raher han o analyss irraional forecas/recommendaion for Inerne socks. If analyss forecas for Inerne socks in he same manner as hey forecas for oher socks, here is no reason o blame hem for heir opimisic forecass for Inerne socks. In order o prove ha analyss were accounable for he formaion of he Inerne Bubble, one 9

10 needs o show ha (1) afer conrolling for he IPO effecs, analyss forecass were relaively more opimisic for Inerne socks han for oher socks, and (2) heir relaive opimism for Inerne socks indeed inflaed he prices of hese socks. 2.3 Hypohesis Developmen Taking ino consideraion analyss over-opimism for all socks in general, I invesigae he exen of heir relaive opimism for Inerne socks hrough H1: during he Inerne Bubble period ( ), analyss made relaively more opimisic sock recommendaions for Inerne socks han for oher socks. Furhermore, I invesigae he impac of heir relaive opimism hrough H2 - analyss opimisic recommendaions for Inerne socks inflaed he prices of hese socks. 3. Research Design 3.1 Sample Selecion The sample covers 2826 firms ha wen public beween January 1, 1996 and Augus 15, I obained my iniial lis of firms from research done by Maureen McNichols, Pa O Brien, and Jerry Sun. They idenified he 2826 IPO firms from he Securiies Daa Corporaion (SDC) daabase, and classified hem ino hree groups - Inerne, non-ne high-ech (NNHT), and non-ech - using he following process: 1. Firs, hey divide he sample of 2826 IPO firms ino wo groups, high-ech and non-ech, according o SDC s indusrial classificaion. Specifically, hey idenify high-ech firms by selecing he IPO firms wih enries in SDC s high-ech indusry field. This resuls in 1492 high-ech firms and 1334 nonech firms. 2. They furher divide he 1492 high-ech IPO firms ino wo sub-groups: 10

11 Inerne firms (Inerne) and non-inerne high-ech firms (NNHT). They idenify Inerne firms by searching hrough he high-ech indusry and long business descripion fields of he SDC daabase and selecing hose firms ha have he keyword Inerne in eiher field. This resuls in 528 Inerne firms and 964 NNHT firms. In his sudy, I use heir indusry secor classificaion and compare analyss opimism across he indusry secors. This design allows for he examinaion of analyss relaive opimism for Inerne socks in paricular while conrolling heir over-opimism for socks in general and new issues in paricular. I also allows me o compare he Inerne and NNHT secors o furher invesigae wheher analyss relaive opimism (if i exised) was Inerne-specific or exended o all high-ech socks. Inerne socks are relaively new compared o socks in radiional indusries, hus randomly selecing Inerne and non-inerne socks will resul in a relaively young Inerne group and a relaively old non-inerne group. Dispariies in firm age have implicaions for analyss opimism: he IPO lieraure suggess ha analyss are more opimisic abou a sock s fuure prospecs around he sock s iniial public offering (e.g., Loughran and Rier, 2002; Rajan and Servaes, 1997). To eliminae he difference in firm age, I examine only IPO socks from 1996 o 2000 for all hree secor groups o ensure ha he secor groups are comparable in age. In addiion, ime-period-specific economic condiions and shocks can influence a firm s performance. Drawing IPOs from he same ime period for he Inerne and non-inerne groups provides a conrol for his imeperiod-specific effec. 11

12 3.2 Daa Collecion I collec analyss recommendaions, sock reurns, and ROE daa for he 2826 IPO socks from Firs Call, CRSP and Compusa respecively. I use he socks perm numbers - he unique idenifier for socks in he CRSP daabase - o mach firms across differen daabases. For each IPO sock, I obain is recommendaion daa for he ime period January 1, 1996 o December 31, 2000 from Firs Call. This resuls in 62,703 recommendaions, each of which represens a broker assigned recommendaion raing o a sock a a poin in ime beween January 1, 1996 and December 31, For each recommendaion, I obain he sock s daily sock reurns from he CRSP daabase and he socks reurn on equiy (ROE) via Research Insigh using he Compusa daabase. The sample collecion saisics are shown in Table 1. Using he final sample of 48,853 recommendaions, I es he hypoheses hrough he regression models discussed in he nex wo secions. 4. The Logisic Regression Tess for H1 4.1 The Logisic Model This sudy uses a ime series logisic regression model o examine analyss relaive opimism oward Inerne socks during he Inerne Bubble period. The model is specified as follows: Recommendaioni, = h( 0 1Inerne j 2 5IPOyear * NNHT j 6 year 7S ROE wheres = {non - ech, NNHT, Inerne} NNHT j IPOyear 3 Sockre I define he variables used in his regression as follows: IPOyear 4 * Inerne Q 1, Q 4 8S 366, 1 9S 1, 366 i, i, Sockre Regression 1 j e 12

13 Recommendaion i, = 1 if analys i assigned a raing of eiher 1 or 2 (srong buy or buy) o sock j a ime ; 0 if he raing assigned was 3, 4, or 5 (hold, sell, or srong sell) Inerne j = NNHT j = IPO_YR y = Year y = 1 if firm j is an Inerne firm; 0 oherwise 1 if firm j is a non-inerne high-ech sock; 0 oherwise 1 if he recommendaion is made wihin one year of he sock s iniial public offering dae; 0 oherwise he calendar year in which he recommendaion is made Sockre -366,-1 = firm j s annualized cumulaive sock reurn for is available rading days in he year prior o he recommendaion dae. Sockre -1,-366 = firm j s annualized cumulaive sock reurn for is available rading days in he year afer he recommendaion dae. ROE j, Q 1, Q 4 = sock j s cumulaive ROE for he four quarers (Q1 o Q4) following he recommendaion quarer (Q0) In his model I use analyss recommendaions as a measure of analyss opimism. Sock raings are discree inegers ranging from 1 o 5. To make analyss recommendaions a more disincive and represenaive measure of analyss opimism, in his sudy I define raings 1 and 2 (srong buy and buy) as favourable and raings 3 hrough 5 (hold, sell and srong sell) as unfavourable. Thus, he dependen variable recommendaion can ake one of wo values, favourable and unfavourable, represened as 1 and 0 respecively. Aside from analyss recommendaions, prior research has also used earning forecas error as a measure for analyss opimism (e.g. Liu and Song, 2000). This measure is inappropriae for use in his sudy because he magniudes of earnings and earnings forecas errors are no comparable across he hree indusry secors. On he oher hand, regardless of he differences in he scope of earnings and earnings forecas errors, analyss recommendaions reflec heir opinions and beliefs abou paricular socks. They 13

14 provide a uniformed scale on which o compare socks in differen secors. Wih non-ech as he base group, he wo indusry secor indicaor variables, Inerne j and NNHT examine wheher analyss are relaively more opimisic abou Inerne/NNHT socks han abou non-ech socks. I hus capures he impac of he indusry secor on an analyss opimism in heir sock recommendaions. This serves as a es for H1, wheher financial analyss are relaively more opimisic abou Inerne socks han abou oher socks. As shown in Table 2 and Figures 2 and 3, he sample IPOs are no disribued evenly hroughou he sampling period; over 50% of he non-inerne IPOs ook place in 1996 and 1997, while over 60% of he Inerne IPOs ook place in 1999 and Exising IPO lieraure suggess ha analyss make more opimisic recommendaions around socks equiy offerings (e.g. Womack, 1996; Lin and McNichols, 1998). To ensure ha he differing disribuions of Inerne and non-inerne IPOs and he resuling differences in he iming of IPO effecs among he indusry secors do no confound he resul, I include an IPO_YR indicaor variable o capure he effec of an IPO year on analyss opimism. The ineracion of he Indusry secor variables wih he IPO indicaion variable enables esing of poenial secor differences in he IPO effec. Also included in he regression is a year variable (indicaing he year in which he recommendaion is made). I use his variable o invesigae wheher he exen of analyss relaive opimism differs across years hroughou he Inerne Bubble period. Pas lieraure suggess ha, while making forecass and recommendaions for socks, analyss observe sock price behavior and learn from invesors decisions; hey primarily reac o changes in marke values raher han cause hem (Amir e al., 1999). I 14

15 use a pas performance variable o ensure ha any effec of socks pas performance on curren recommendaions did no confound indusry secor-based opimism (as capured by he secor indicaor variables). To invesigae any poenial secor differences, I use a secor-based pas performance variables - 8S Sockre --366,-1, where S = {non - ech, NNHT, Inerne} - o separaely measure he exen of his effec in each indusry secor. I define pas performance by he annualized cumulaive sock reurn in he year prior o he recommendaion dae. Since he reurns daa for he enire year prior o he recommendaion dae are unavailable for recommendaions ha are made wihin a sock s IPO year (he sock does no have one year of reurns hisory), I annualize he cumulaive reurn for he available rading days o make he pas performance measure applicable o hose recommendaions. The formula for he annualized cumulaive reurn is shown as follows: j= ( 1) (1 rj ) j= ( 366) Cumulaive _ reurn = [ * 250] 1 n According o his formula, I firs accumulae he sock s daily reurns ( r j ) for all available rading days in he year prior o he recommendaion dae (). Then I divide his cumulaive reurn by he number of rading days accumulaed (n), and muliply he resul by 250 (he approximae number of oal rading days in a year). Finally, I deduc one from he resul o obain he annualized reurn - he measure for pas performance. Prior research suggess ha analyss recommendaions and forecass have impac on and are impaced by he socks fuure prices (Amir e al., 1999). I include a secorbased fuure sock performance variable (Sockre -1,-366) in he regression as a conrol for his effec. Keeping consisency beween he pas and fuure sock performance 15

16 variables, I define fuure sock performance by he annualized cumulaive sock reurns in he year afer he recommendaion dae and calculae i in he same manner as I calculae he pas sock performance measure. Besides he above causes of recommendaion opimism (i.e. IPO year, socks pas and fuure sock performance), an analys may hold a relaively more opimisic view for a paricular sock because he/she foresees he sock s superior fuure earnings poenial. I include a fuure earnings performance variable ( ROE Q 1, Q 4 ) o measure he exen o which analyss opimism for a paricular sock can be explained by he sock s superior fuure earnings poenial. I measure a sock s fuure earnings performance by he firm s cumulaive ROE for he four quarers (Q1 o Q4) following he recommendaion quarer (Q0). Again, I define he ROE measure for each indusry secor o allow for poenial secor differences. The resuls for Regression 1 are shown in Table 3. The esimaed inercep erm (ß 0 ) is he esimaed probabiliy ha a non-ech sock obains a favorable sock recommendaion (raing 1 or 2) in he absence of he effecs of oher explanaory variables. The esimaed coefficien of he Inerne secor indicaor variable (ß 1 ) is quaniaively large and saisically significan. I suggess ha, holding consan he oher facors of analyss opimism (i.e. IPO year and he sock s pas and fuure performance), being in he Inerne secor increases a sock s probabiliy of obaining a favorable recommendaion when compared o socks in he non-ech secor. The coefficien of he NNHT secor indicaor variable (ß 2 ) is saisically insignifican a he 5% significance level. This indicaes ha, compared o being in he non-ech secor, being in he NNHT secor does no significanly increase a sock s probabiliy of obaining a 16

17 more favorable sock recommendaion. The esimaed coefficien ß 1 is greaer han ß 2, suggesing ha analyss recommendaions for he Inerne socks are more opimisic han hose for he NNHT socks. To invesigae wheher his difference is saisically significan, I perform he following es: Tes 1: Ho: There is no significan difference in he exen of analyss opimism beween he Inerne and NNHT secors. Ha: The exen of analyss opimism differs significanly beween he Inerne and NNHT secors. Alernaively, Ho: ß 1 = ß 2 Ha: ß 1 4 ß 2 The es resul is shown as es 1 in Table 4. On he basis of he small p-value (<0.0001) of he chi-square saisics, I rejec he null hypohesis and conclude ha analyss made significanly more opimisic recommendaions for Inerne socks han for NNHT socks during he Inerne Bubble period. This, combined wih he facs ha analyss are significanly more opimisic abou Inerne socks han abou non-ech socks and ha heir opimism does no differ significanly beween he NNHT and nonech socks, provides evidence ha analyss relaive opimism is Inerne sock-specific and does no apply o non-inerne socks in he high-ech secor. This provides suppor for H1 - during he Inerne Bubble period, analyss made relaively more opimisic sock recommendaions for he Inerne socks han for non-inerne socks. The IPO_YR indicaor variable capures he IPO effec. The posiive and saisically significan coefficien of his variable (ß 3 ) suggess ha analyss are on 17

18 average more opimisic in heir sock raings during a firm s IPO years han during is non-ipo years. This confirms he IPO effec described in he IPO lieraure. The saisically insignifican coefficiens of he ineracion of he IPO year indicaor variable and he secor indicaor variables (ß 4 and ß 5 ) sugges ha he IPO effec does no differ significanly across indusry secors. The coefficiens of he fuure earnings performance variables, ß 7S, measure o wha exen a firm s fuure earnings performance explains analyss recommendaion opimism for socks in he non-ech, Inerne and NNHT secors respecively. The saisically significan coefficien of he fuure earnings performance measure for he non-ech secor, ß 7S, where S = non-ech, suggess ha here is a posiive correlaion beween analyss recommendaions and he fuure earnings performance of non-ech socks. I indicaes ha analyss recommendaions for non-ech socks are a predicor of he sock s fuure earnings poenial. The esimaed coefficiens of he fuure earnings performance variables for he Inerne and NNHT secors are saisically insignifican a he 5% level. This provides evidence ha, on average, analyss recommendaions for Inerne and NNHT socks are no predicive of he sock s fuure earnings poenial. I also lends some credence o he financial press s claims ha analyss made irraional and incredible forecass and recommendaions for Inerne socks during he Inerne Bubble period. The quaniaively large and saisically significan coefficiens of he pas performance variables, ß 8S provide evidence ha analyss curren recommendaions for socks in all hree secors are posiively correlaed wih he sock s pas performance. Tha is, socks wih superior pas reurns are more likely o receive favorable sock recommendaions. The coefficien of pas performance variable for he non-ech secor is 18

19 greaer han ha for he NNHT secor, which in urn is greaer han ha for he Inerne secor. To furher invesigae wheher he secor differences are saisically significan, I carry ou he following ess: Tes 2: Ho: There is no significan difference in he effec of pas performance on analyss recommendaions beween he Inerne and non-ech secors. Ha: There is a significan difference in he effec of pas performance on recommendaions beween he Inerne and non-ech secors. Alernaively, Ho: ß 8S, where S=Inerne = ß 8S, where S = non-ech Ha: ß 8S, where S=Inerne 4 ß 8S, where S = non-ech Tes 3: Ho: There is no significan difference in he effec of pas performance on recommendaions beween he Inerne and NNHT secors. Ha: There is a significan difference in he effec of pas performance on recommendaions beween he Inerne and NNHT secors. Alernaively, Ho: ß 8S, where S=Inerne = ß 8S, where S = NNHT Ha: ß 8S, where S=Inerne 4 ß 8S, where S = NNHT The es resuls are shown as Tes 2 and Tes 3 (he second and hird iems) in Table 4. The small p-values of he chi-square saisics lead o he rejecion of he null hypoheses for boh ess. Thus, I conclude ha pas performance has a significanly greaer impac on analyss recommendaions for he non-ech socks han for he NNHT socks, and ha i has a significanly greaer impac on analyss recommendaions for he NNHT socks han for he Inerne socks. The coefficiens of he fuure sock reurn variables measure o wha exen analyss recommendaion opimism can be explained by a sock s fuure reurns. I hus conrols 19

20 for he effec of he socks fuure performance as an explanaion for analyss recommendaion opimism. The esimaed coefficiens for he fuure reurns measure for all secors are saisically insignifican. This suggess ha, on average, he superior/poor fuure sock performance of he socks did no drive analyss opimism/pessimism in heir recommendaions for hese socks. This lends suppor o Amir s claim ha, in general, analyss reac o price changes raher han cause hem. One possible alernaive explanaion for he observed resul is ha, because ROE and sock reurns are highly correlaed, inclusion of boh variables caused a co-lineariy problem and confounded he resul. To address his concern, I run he regression including only he fuure sock reurn measures. The new regression is specified as follows: Re commendaioni, = h( 0 1Inerne j 2NNHT 5IPOyear * NNHT j 6 year 7S Sockre where S = {non - ech, NNHT, Inerne} j IPOyear 3 366, 1 8S Sockre IPOyear 4 1, 366 e i, i, * Inerne Regression 2 The resuls are shown in Table 5. As shown in he able, he coefficien for he Inerne secor indicaor variable is sill saisically significan and ha for he NNHT is sill insignifican; he esimaed coefficiens for he fuure sock reurn measure for he Inerne socks is sill negaive and significan, while hose for NNHT and non-ech are insignifican even when he ROE measures are aken ou of he regression. This suggess ha he insignifican coefficiens of he fuure performance variables are no caused by including boh fuure earnings and sock reurn variables. j 4.2 The Yearly Analysis The esimaed coefficien for he year variable (ß 6 ) in Regression 1 is saisically 20

21 significan, suggesing a significan difference in he exen of opimism across years. To furher invesigae his difference, I re-run regression 1 for each year. To demonsrae he difference in he exen of he secor-based recommendaion opimism across year, I summarize he coefficiens and heir corresponding P-values for he indusry secor indicaor variables of he yearly regressions in Table 6. Coefficiens of he inercep erms measure he exen of analyss recommendaion opimism for he base group socks in he nontech secor. As shown in he able, analyss are opimisic in heir sock recommendaions for socks in he nontech secor hroughou he sampling period (1996 o 2000). Using nontech as he base group, he coefficiens of he Inerne and NNHT secor indicaor variables measure he exen of analyss relaive opimism for hese socks in he Inerne and NNHT secors. As shown in he able, in 1996, 1997 and 1998, analyss recommendaions for he Inerne and NNHT socks are no saisically significanly differen from hose for he non-ech socks. In 1999 and 2000, heir recommendaions for he Inerne socks are significanly more opimisic han hose for nontech socks, while heir recommendaions for he NNHT socks are significanly more pessimisic han hose for he nontech socks. To conclude, in he earlier years of he Inerne Bubble period ( ), analyss did no repor more opimisically for he high ech socks han oher socks; heir relaive opimism for Inerne socks is concenraed in he laer years of he Bubble period (1999 and 2000), and his opimism did no apply o he oher non-inerne high-ech socks. The resuls of boh he ime series and yearly regression analyses provide srong evidence for analyss relaive opimism for he Inerne socks during he Inerne Bubble period. However, o hold financial analyss responsible for creaing he Inerne Bubble, 21

22 one also has o prove ha heir relaive opimism for Inerne socks indeed inflaed he prices of hese socks during he Bubble period. The nex secion ess he validiy of his claim. 5. The OLS analysis Tes for H2 I use an Ordinary Linear Regression (OLS) o es H2 - financial analyss sock recommendaions inflaed he prices of Inerne socks. The regression model is specified as follows: Sockre 1, 366 = 0 1Inerne j 2 NNHT j 3 year where S = {non - ech, NNHT, Inerne} IPOyear raing 4 5S i, i, e Wih fuure sock reurn ( Sockre 1, 366 ) as he dependen variable and indusry secor-based recommendaion raings ( raing 5S,where S = {non-ech, NNHT, Inerne) as he es variables, his regression examines wheher analyss sock recommendaions for socks in each indusry secor drove he fuure prices of hese socks during he Inerne Bubble period. During he Inerne Bubble period, high-ech socks over-performed radiional socks in erms of sock reurns. To conrol he effec of indusry secor on sock reurns, I include in he regression secor indicaor variables, Inerne and NNHT. Since imeperiod-specific economic condiions and shocks can influence a firm s performance, I include a year variable as a conrol for he effec of economic condiions on sock reurns. Also, he IPO lieraure suggess ha IPO socks on average are under-priced (e.g. Krigman e al., 1999). To accoun for poenial IPO under-pricing, I include in he regression an IPOyear indicaor variable. 22

23 The resul of his regression is shown in Table 7. The inercep erm measures he effec of being in he nontech secor on he sock s fuure reurns in absence of he effecs of any oher explanaory variables. Wih he nontech secor as he base group, he wo indusry secor indicaor variables (Inerne and NNHT) es wheher being in a paricular secor improves he socks reurn performance. The posiive and saisically significan coefficiens of hese wo secor indicaor variables indicae ha, compared o socks in he non-tech secor, Inerne and NNHT socks experienced higher reurns during he Inerne Bubble period. The negaive and saisically significan coefficien of he IPOyear indicaor variable suggess ha in general socks have lower reurns in heir IPO years han in non-ipo years. This confirms o IPO under-pricing described in he IPO lieraure. The coefficiens of he hree secor-based recommendaion variables (, where S {non - ech, NNHT, Inerne} ) measure he effec of analyss 5 S = recommendaion raings on he socks fuure reurns. The coefficiens for all hree variables are quaniaively small and saisically insignifican (a he 5% level). This suggess ha analyss recommendaions for socks in all secors have no significan impac on he socks fuure reurns. This provides evidence agains he claim ha analyss opimisic recommendaions for he Inerne socks inflaed he prices of hese socks. Based on his evidence, I rejec H2 and conclude ha financial analyss opimisic recommendaions for he Inerne socks did no drive he prices of hese socks during he Inerne Bubble period. 23

24 6. Conclusion In his paper I invesigae he role financial analyss played in he 2000 Inerne Bubble by empirically examining he exen and impac of financial analyss relaive opimism for he Inerne socks hroughou and in each year of he Inerne Bubble period ( ). The Logisic regression analysis suggess ha during he Inerne Bubble period analyss made relaively more opimisic recommendaions for Inerne socks han for oher socks; his relaive opimism was Inerne sock-specific and did no apply o he non-inerne socks in he high-ech secor; he degree of analyss relaive opimism for Inerne socks differs significanly across years - in he earlier years of he Inerne Bubble (1996 o 1998) analyss were no significanly more opimisic for Inerne socks han for oher socks; however, in he laer years of he Bubble period (1999 and 2000) hey were significanly more opimisic for Inerne socks han for oher socks. These resuls provide srong suppor o he claim ha analyss are relaively more opimisic for Inerne socks han for non-inerne socks. The OLS regression analysis shows ha analyss recommendaions have no significan impac on he fuure sock prices of he Inerne socks. This provides evidence agains he claim ha financial analyss opimisic sock recommendaions inflaed he prices of hese socks and caused he Inerne Bubble. Combining he resuls from boh analyses, I conclude ha, as he financial press speculaed, financial analyss indeed held a relaively more opimisic view for he Inerne socks during he Inerne Bubble period; however, heir relaive opimisic sock recommendaions for he Inerne socks did no 24

25 inflae he prices of hese socks and hus hey should no be blamed enirely for creaing he Inerne Bubble. 25

26 References Amir, E., Lev, B., and Sougiannis, T Wha Value Analyss? Working Paper, <hp://ssrn.com/absrac=193428> Barefield, R., Comiskey, E., The accuracy of analyss forecass of earnings per share. Journal of Business Research 3, Bradley, D. J., Jordan, B. D., and Rier, J. R. 2003, The Quie Period Goes Ou wih a Bang. Journal of Finance, 58, Cooper, A., C. Woo, and W. Dunkelberg Enrepreneurs perceived chances for success. Journal of Business Venuring, 50(5), Cornell, B. and Liu, Q The Paren Company Puzzle: When is he Whole Worh Less han One of he Pars? Working Paper, Anderson Graduae School of Managemen, UCLA. Darlin, D Picking a Loser: Young Analys Defied Expers and Foresaw Baldwin-Unied s Ills. Wall Sree Journal: 1ff. Dechow, P., Huon, A., and Sloan, R The Relaion beween Analyss Long-Term Earnings Growh Forecass and Sock Price Performance Following Equiy Offerings. Working Paper, Universiy of Pennsylvania and Harvard Universiy. Dirks, R. and Cross., I The Grea Wall Sree Scandal. New York: McGraw-Hill. Fransic, J. & Phyilbrick, D Analyss Decisions as Producs of a Muli-Task Environmen. Journal of Accouning Research: Hand, J Profis, losses and he non-linear pricing of Inerne socks, working paper, Universiy of Norh Carolina. Hiriam H. The sock marke s glory days are over, a leas for now. The Philadelphia Inquirer. December 31, 2000 Sunday D Ediion, Pg. C01 Kohari, S.P., Capial markes research in accouning. Journal of Accouning and Economics 31 (2001) Krigman, L., W. Shaw, and K. Womack The Persisence of IPO Mispricing and he Predicive Power of Flipping. Journal of Finance

27 Lim, T., Are analyss forecass opimisically biased? Working Paper, Darmouh Universiy. Lin, H., and M. McNichols Analys coverage of iniial public offering firms. Working Paper, Sandford Universiy. Liu, Q. and Song., F The Rise and Fall of Inerne Socks: Should Financial Analyss be Blamed? Working Paper, School of Economics, Universiy of Hong Kong Loughran, T. and Rier, J. R. 2002, Why Don Issuers Ge Upse Abou Leaving Money on he Table in IPOs? Review of Financial Sudies, 15, McNichols, M. and O Brien, P Self-selecion and analys coverage. Journal of Accouning Research 35 (Supplemen): Michaely, R. and Womack, K. L Conflic of Ineres and he Credibiliy of Underwrier Analys Recommendaions, Review of Financial Sudies, 12, O Brien, P Analyss forecass as earnings expecaions. Journal of Accouning and Economics 10, Rajan, R. and Servaes, H Analys Following of Iniial Public Offerings, Journal of Finance, 52, Trueman, B., Wong, M., and Zhang, X The Eyeballs have i: Searching for he Value in Inerne Socks. forhcoming, Journal of Accouning Research. 27

28 Figure 1 Major Indexes (01/31/99 =100) 28

29 Figure 2: Disribuion of Inerne IPOs Disribuion of Inerne IPOs by year % % % % % Figure 3: Disribuion of non-inerne IPOs Disribuion of non-inerne IPOs by year % % % % %

30 Table 1 Sample Selecion Saisics The IPO socks are obained from he SDC daabase. The able summarizes he number of socks wen public during , he number of socks wih missing daa, he number of recommendaions for he remaining IPO socks, he number of recommendaions wih missing daa, and he number of he remaining recommendaions. Saisics Cumulaive Saisics Number of IPO socks from he SDC daabase 2,826 Number of IPOs missing perm number in CRSP 20 Number of remaining IPO socks 2,806 Number of recommendaions during ,690 Number of recommendaions missing ROE daa 13,810 Number of Remaining recommendaions 48,880 Number of recommendaions missing sock reurn daa 27 Final sample of recommendaions 48,853 Table 2: Disribuion of IPOs by Year ( ) This able displays he number and percenage disribuion of IPO socks in each indusry secor across he ime period 1996 o Number of IPOs Percenage Year Inerne nontech NNHT Inerne nontech NNHT % 36.7% 34.4% % 28.7% 22.0% % 18.7% 10.4% % 11.4% 16.1% % 4.5% 17.1% Toal % 100.0% 100.0% 30

31 Table 3: Logisic Regression of Recommendaions on Indusry Secors (Regression 1) In his able, I repor he resuls of he logisic regression of recommendaions on indusry secors. The sample covers 48,853 recommendaions for he 2806 IPO socks for he ime period 1996 o The regression is specified as follows: Re commendaioni, = h( 0 1Inerne j 2 NNHT j 3IPOyear 4IPOyear * Inerne j 5IPOyear * NNHT j 6 year 7S ROE Q 1, Q 4 8S Sockre 366, 1 9S Sockre 1, 366 where S = {non - ech, NNHT, Inerne} Explanaory Variables (Coefficiens) Esimaed Coefficien i, i, Sandard Error Wald Chi- Square Pr>ChiSq Inercep (ß 0 ) Inerne (ß 1 ) NNHT (ß 2 ) IPOyear (ß 3 ) inerne*ipo_yr (ß 4 ) nnh*ipo_yr (ß 5 ) Year (ß 6 ) (S=non-ech) ROE 7 S Q 1, Q 4 (S=Inerne) ROE 7 S Q 1, Q 4 (S=NNHT) ROE 7 S Q 1, Q 4 8s 366, 1 Sockre (S=non-Tech) s 366, 1 Sockre (S=Inerne) s 366, 1 Sockre (S=NNHT) Sockre 9 S 1, Sockre 9 S 1, 366 (S=Inerne) Sockre 9 S 1, 366 (S=NNHT) e 31

32 Table 4: Tess for Secor Differences This able summarizes resuls of he following hree ess. Tes 1 examines wheher he exen of analyss opimism for he Inerne and NNHT secors differ significanly. Tes 2 examines wheher he effec of pas performance on analyss recommendaions for he Inerne and non-ech secors are significanly differen. Tes 3 examines wheher he effec of pas performance on analyss recommendaions for he Inerne and NNHT secors are significanly differen. Wald Chi-Square DF Pr>ChiSq Tes 1: nnh = inerne Tes 2 : ß 6S, where S = Inerne = ß 6S, where S = nontech ; Tes 3: ß 6S, where S = Inerne = ß 6S, where S = NNHT Table 5: Logisic Regression of Recommendaion on Indusry Secors (Regression 2) This able repors resul of running he logisic regression of recommendaions on indusry secors (Regression 1), while excluding he ROE measures. Explanaory Variables (Coefficiens) Esimaed Coefficien Sandar d Error Wald Chi- Square Pr> ChiSq Inercep (ß 0 ) Inerne (ß 1 ) NNHT (ß 2 ) IPOyear (ß 3 ) inerne*ipo_yr (ß 4 ) nnh*ipo_yr (ß 5 ) Year (ß 6 ) s 366, 1 Sockre (S=non-Tech) s 366, 1 Sockre (S=Inerne) s 366, 1 Sockre (S=NNHT) Sockre 8 S 1, Sockre 8 S 1, 366 (S=Inerne) Sockre 8 S 1, 366 (S=NNHT)

33 Table 6: Summary of he Yearly Logisic Regression Analyses ( ) The logisic regression of recommendaion on indusry secors (Regression 1) is run separaely for each year This able summarizes he esimaed coefficiens (ß 0, ß 1, and ß 3 ) and he corresponding p-values of he indusry secor variables (nontech, Inerne, and NNHT) in each yearly analysis. Variable Inercep (ß 0 ) (0.083) (0.001) (0.001) (0.001) (0.001) Inerne (ß 1 ) (0.703) (0.522) (0.171) (0.007) (0.001) NNHT (ß 2 ) (0.947) (0.947) (0.188) (0.001) (0.202) Table 7: Ordinary Leas Square Regression of Fuure Sock Reurns on Secor-Based Recommendaion Raings This able repors he OLS resul of regressing fuure sock reurns on analyss sock recommendaions. The fuure sock reurns are calculaed by annualizing he cumulaive sock reurns for he one year following he recommendaion dae. The recommendaion variable is defined for each indusry secor o allow for poenial secor differences. The model is specified as follows: Sockre 1, 366 = 0 1Inerne j 2NNHT j 3 year 4IPOyear raing e where S = {non - ech, NNHT, Inerne} 5S i, i, Explanaory Variables (Coefficiens) Parameer Esimae Sandar d Error T value Pr> ½ *Pvalue Inercep (ß 0 ) Inerne (ß 1 ) NNHT (ß 2 ) year (ß 3 ) IPOyear (ß 4 ) S Raing (S=Inerne) (ß 5 ) S Raing (S=NNHT) (ß 5 ) S Raing (S=nonTech) (ß 5 )