A copy can be downloaded for personal non-commercial research or study, without prior permission or charge

Transcription

1 Sganos, A. (2013) Google attenton and target prce run ups. Internatonal Revew of Fnancal Analyss. ISSN Copyrght 2012 Elsever A copy can be downloaded for personal non-commercal research or study, wthout pror permsson or charge The content must not be changed n any way or reproduced n any format or medum wthout the formal permsson of the copyrght holder(s) When referrng to ths work, full bblographc detals must be gven Deposted on: 18 February 2013 Enlghten Research publcatons by members of the Unversty of Glasgow

2 Google attenton and target prce run ups Antonos Sganos Unversty of Glasgow Dvson of Accountng and Fnance, Adam Smth Busness School, Unversty of Glasgow, West Quadrangle, Man Buldng, Unversty Avenue, Glasgow, G12 8QQ, Scotland, e-mal: Acknowledgement I am grateful to Gllan MacIver for her support wth data collecton, to FSA staff for an nsghtful dscusson on the topc, and to two anonymous referees for valuable suggestons. 1

3 Google attenton and target prce run ups ABSTRACT We explore the ncrease n the share prces of target frms before ther merger announcements. We use a novelty Google search volume to proxy the market expectaton hypothess accordng to whch frms wth an abnormal upward change n Google searches are dentfed as frms wth potental merger actvty. We fnd that Google ndcators can explan a larger percentage of the prce ncrease n target frms before ther mergers than the Fnancal Tmes. However even the Google proxy of the market expectaton hypothess can only explan at best 36 percent of the target prce run ups. Keywords: Target prce run ups, mergers, market antcpaton, Google search volume JEL classfcaton: G14; G34 2

4 Google attenton and target prce run ups 1. Introducton A number of academc studes have reported that share prces of target frms do ncrease sgnfcantly pror to ther merger announcement and have developed two hypotheses to explan such a pattern. Accordng to the nsder tradng hypothess (Keown and Pnkerton, 1981), staff from the target, bddng, or fnancal nsttuton that organzed the transacton trade or even pass such nformaton on to relatve members. Accordng to the alternatve market expectaton hypothess (Jensen and Ruback, 1983), nvestors, based on publcly avalable nformaton, manage to predct target frms pror to ther merger announcements. Ths paper focuses on the latter hypothess explorng whether the target prce run ups are drven by publc nformaton. Pror studes have used meda coverage to proxy the market expectaton hypothess, wth nvestors beng able to predct target frms as long as such nformaton was documented n the meda. Early studes n the feld (e.g., Pound and Zeckhauser, 1990; Zvney et al., 1996) have focused on the newspaper coverage of a partcular column, such as the columns Heard on the Street and/or Abreast of the Market, wth more recent studes (e.g., Kng, 2009) ncorporatng a wder coverage of artcles wth the assstance of databases such as Factva. The majorty of those studes have concluded that meda coverage can only explan part of the ncrease n target share prces pror to ther merger announcements. Wthn the UK lterature, Holland and Hodgknson (1994) explore 86 target frms from 1988 to 1989 and Sganos and Papa (2012) 1,059 frms between 1998 and Wthn Holland and Hodgknson s lmted sample, rumors covered by Fnancal Tmes (FT) drve to a large extent the UK target prce run ups. Sganos and Papa report that n lne wth nternatonal lterature, FT coverage of rumors can only explan a small percentage of 3

5 the upwards UK target pattern. However, pror studes have not captured all publcly avalable nformaton; as an example, none of the pror UK/nternatonal studes n the feld has ncorporated nvestors dscussons on onlne stes such as Hotcopper.com.au, though Clarkson et al. (2006) and Chou et al. (2010) have found that such merger rumors have a sgnfcant mpact on frms share returns. Therefore, pror studes conclusons may be based due to the lmted news coverage. Based on the dffculty of capturng all avalable publc nformaton, we explore an alternatve approach to proxy the market expectaton hypothess by usng the volume of Google searches for target frms. Google s the most wdely used web search engne and the only search ste that offers hstorc searchng volume data approprate for academc purposes. 1 If nvestors encounter a rumor of a potental merger, most nvestors may use Google to search for further nformaton on the target company before proceedng wth a transacton; therefore, frms featured n a rumor are expected to experence an abnormal ncrease n Google search actvty. A few recent studes have reported the sgnfcance of Google searches as a measure of nvestor attenton. Da et al. (2011) explore the best proxy of nvestor attenton n US frms and fnd that Google searches capture nvestor attenton earler than exstng proxes, such as newspaper coverage, and Bank et al. (2011) support the sgnfcance of Google search volume as a proxy of nvestor attenton n German stocks. Other recent studes have also shown the sgnfcance of Google searches wthn alternatve felds n fnance. Da et al. (2012) report that Google searches are value relevant and have the ablty to predct frms revenue surprses, and Drake et al. (2011) report that Google searches are related to frms prce and tradng volume levels before and on the earnngs announcement day, wth frms wth hgh Google actvty pror to the announcement experencng a smaller prce and volume response on the announcement day. 1 For a bref revew of Google, study (last accessed September 2012). 4

6 We study the Google search volume of target frms wthn the merger context to explore whether the mergers were expected by nvestors. We frst explore whether Google search volume can predct mergers before such rumors are reported n FT and, second, whether Google attenton can explan the target prce run ups pattern. As an example, Fgure 1 shows daly Google search volume for RHM plc between October and December 2006; Premer Foods plc acqured RHM plc on 4 th December We fnd an ncrease n the volume of RHM s Google search actvty a few days pror to the merger announcement: Google attenton was 0.13 on 29/11/2006, 0.19 on 30/11/2006, 0.39 on 1/12/2006, 0.59 on 2/12/2006, 0.80 on 3/12/2006, and 1 on 4/12/2006, before movng back to normal levels of Google attenton. 2 Between October and December 2006, we have only dentfed two FT artcles that document a potental merger deal for RHM, publshed on 2/12/2006 (Wggns and Hume, 2/12/2006) and on 3/12/2006 (Wggns, 3/12/2006). There s, therefore, a sgn that nvestors were searchng for nformaton on RHM plc earler than FT covered potental merger actvty. To test our argument, we manually download daly Google actvty for 340 UK target frms between March 2004 and December We adopt the outler lterature to dentfy abnormal upward changes n Google searches by usng the boxplot method (Tukey, 1977), whch makes no dstrbutonal assumptons. Followng an event study analyss, we estmate excess returns of target frms before ther merger announcement date and before the frst date that abnormal Google actvty was sgnaled. We fnd that Google ndcators tend to offer a takeover sgnal a few days earler than FT, and we therefore fnd that Google ndcators explan a larger percentage of the prce ncrease n target frms than a conventonal FT 2 Notce that Google search data are gven at a relatve value to the total searches n the sample perod requested that ranges between 0 and 1, where 1 ndcates the day wth the maxmum number of searches. Also notce that Google search volumes may slghtly change when collected at dfferent ponts n tme, snce Google calculates the values from a subset of the full archve to ncrease the response speed. In lne wth Da et al. (2011), we download results for a few frms wthn alternatve tmes and fnd that the correlaton of the data s above 0.95; we therefore conclude that our results are not drven by such approxmatons. 5

7 coverage proxy. Nevertheless, even after estmatng excess returns before the Google merger sgnals, the target prce run ups reman economcally and statstcally sgnfcant, showng that Google ndcators cannot fully explan the prce pattern. We fnd that Google can explan at best merely 36 percent of the target prce run ups. The remander of the paper s structured as follows. The next secton explans the data and methodology used, secton 3 dscusses the emprcal results and secton 4 concludes the study. 2. Data and methodology 2.1 Data collecton We use Thomson OneBanker to have access to all UK target frms wth at least a 50 percent level of acquston between March 2004 and December To be selected n the sample, a target frm should have an avalable Datastream code n Thomson OneBanker (to lnk Thomson OneBanker and Datastream), daly share returns 3 and a tcker symbol n Datastream. In lne wth other studes that have used Google data (e.g., Da et al., 2011), we use the tcker, rather than the name of the frm, to collect data from Google, snce tckers are promnently used by nvestors rather than by consumers nterested n a frm s product. The fnal sample conssts of 430 target frms. In unreported results, we fnd that the ncrease n the share prces of those target frms commences 30 days pror to ther merger announcement, whch sets the sample perod of the study and s well n lne wth the tme frame used n the majorty of studes n the feld (e.g., Holland and Hodgknson, 1994). 3 We use the RI data type that ncorporates dvdend payments n the estmaton of share returns. 6

8 Two man data sources are used to explan the target prce run ups. Frst, we download FT coverage for target frms from NewsBank. 4 In lne wth Dyck et al. (2008) and Ferguson et al. (2011), we focus on FT coverage snce FT s the most nfluental newspaper wth the most credblty among nvestors and s also the most comprehensve for frms fnancal news. Sganos and Papa (2012) and Holland and Hodgknson (1994) are also the only other UK studes n the target prce run ups lterature, and both studes have used FT coverage to proxy news; we therefore focus on FT for comparson purposes. Second, we manually download daly Google actvty for each target frm by usng the Google nsght webste. 5 Notce that Google offers hstorc volume data snce January 2004, whch determnes the sample perod of the study. Out of 430 target frms, 95 frms do not have avalable daly Google data, snce Google reports research volume data only f a search s above a mnmum threshold. We retan all 430 frms n our analyss, snce lmted Google searches for a target frm ndcate that nvestors dd not manage to predct the merger. The Google search data are also gven as a relatve value to the total searches n the sample perod requested, whch ranges between 0 and 1, wth 1 ndcatng the day wth the maxmum number of searches. We use the worldwde selecton to dentfy frms research volume over tme to ncorporate the actvty of nternatonal nvestors and, snce we requre daly frequency, we also set a three-month perod before each frm s merger, whch covers the perod that the target prce run ups pattern s present Determnng abnormal upward change n Google actvty We estmate abnormal Google volume usng two measures for robustness purposes: 4 Notce that the FT data used n ths study have also been used by Sganos and Papa s (2012) study explorng the target prce run ups n the UK between 1998 and The current study nstead focuses on whether Google ndcators can explan the target prce run ups pattern and whether Google can explan a larger percentage of returns than FT. For a more detaled descrpton on FT data, please study the above mentoned paper. 5 (last accessed September 2012). 6 If the length of the search s longer than three months, data are only avalable n a weekly frequency. 7

9 AGoogle ln(1 + Google ) ln(1 + Google ) (1) 1 = t t 1 AGoogle ln(1 + Google ) ln[ Medan(1 + Google ),...,(1 Google )] (2) 2 = t t 41 + t 50 where Google t s the Google actvty of frm on day t that we adjust to a range between 1 and 2 for estmaton purposes. and are estmated daly from -40 days to the day of the merger announcement to capture the target prce run ups. shows the daly change n Google search volume and shows the abnormal daly change above the normal Google actvty for each target frm, as estmated by the medan number of searches about the frm between -41 and -50 days before the merger. Fgure 2 shows the cumulatve abnormal Google searches for both measures from -30 days untl the day the merger was announced. The fgure shows that both and tend to show an ncrease n Google actvty closer to the merger announcement day, wth such upward pattern beng stronger for. The average daly growth of the Google search volume over the last fve days pror to the merger announcement for / s 0.29/0.40 percent, and the correspondng growth on the day of the merger announcement s 0.37 and 1.19 percent, respectvely. To dentfy the frst day that nvestors are aware of a potental merger, we frst use FT coverage and select the frst rumor artcle avalable for each target frm. Snce and are contnuous varables, we follow the outler lterature to determne abnormal upward changes. To determne the most approprate statstc to explore outlers, Table 1 explores the dstrbuton of both measures. We fnd that and are postvely skewed (0.11 and 0.25, respectvely), wth acute peaks (11.13 and 9.41, respectvely), and Kolmogorov Smrnov statstcs show that and do 8

10 not follow normal dstrbuton at the 1 percent level. We therefore use the boxplot method (Tukey, 1977) to dentfy outlers, whch makes no dstrbutonal assumptons and s applcable to data that s not heavly skewed. We dentfy outlers as follows: Outler t > Q + 1.5*( Q3 Q1 ) (3) 3 where Q3 and Q1 are, respectvely, upper and lower quartles for frm over the perod between -40 days and the day of the merger announcement. The frst abnormal upward change n Google actvty for each frm s the frst sgnal of a potental takeover actvty. Table 2 explores the relatonshp between Google and target share returns. Panel A of Table 2 explores whether Google s abnormal upward changes n and have an mpact on share returns on the day that those sgnals were dentfed as well as over the followng days. Notce that excess returns at day t ndcate a smple arthmetc average across all target frms for that day and for robustness reasons, we estmate alternatve rskadjusted excess returns: the market excess returns ( ER M ) 7, the CAPM market-adjusted returns ( ER CAPM ) 8, and the three-factor model adjusted returns ( ER FF 3 ) 9. We fnd that the daly market-adjusted returns ( ER ) for are 1.74, 1.57, 0.81, and 0.25 percent M on day t, t + 1, t + 2, and t + 3 after the takeover sgnal, wth returns beng statstcally sgnfcant at least at the 10 percent level untl two days after the sgnal. These results show that there s a lnk between abnormally hgh Google searches and correspondng share M = Rt RMt, where t 7 ER R s the return of frm n day t, and RMt s the market return (FTSE All Share) n day t. 8 ER ˆ ˆ CAPM = Rt ( a + bm RMt ), where aˆ, b ˆ M coeffcents are estmated over the nterval from -41 to -120 days before the merger announcement (day 0). 9 ER ˆ ˆ ˆ ˆ FF 3 = Rt ( a + bm RMt + bsmbsmbt + bhmlhmlt ), where SMBt and HML t reflect the sze and book/market rsk proxes, respectvely. The factors are estmated n lne wth Fama and French (1993). aˆ, bˆ M, bˆ SMB, bˆ HML coeffcents are estmated over the nterval from -41 to -120 days before the merger announcement (day 0). 9

11 returns; therefore, when we estmate at a later stage returns before such merger sgnals, excess returns are expected to decrease. Panel B of Table 2 further explores the robustness of pror result by testng whether there s n general a postve relatonshp between share returns and Google actvty. We estmate the followng OLS regresson: ER + t = a0 + a1 AGooglet + a2 AGooglet 1 + a3 AGooglet 2 + a4 AGooglet 3 ut (4) where all three measures are used to estmate excess returns ER t ( ER M, ER CAPM and ER FF 3 ) and both Google s volume are employed ( and ). We estmate the above regresson wthn frms wth data avalable between ther ntal takeover sgnal and ther merger announcement day. Notce that all days/data per frm between the merger announcement and the frst merger sgnal are ncorporated nto the regresson. In unreported results, we also estmate the Varance Inflaton Factors (VIFs), to test for potental multcollnearty amongst the explanatory varables, and fnd that most VIFs are slghtly over 1 and the maxmum VIF s equal to These results show that there s no evdence of multcollnearty. Emprcal results to some extent support fndngs reported at Panel A of Table 2, snce we fnd that there s a postve relatonshp between contemporaneous/lagged Google actvty and target share returns when up to two lags are employed. In unreported results, we further reestmate above regresson per day before the merger announcement day and fnd that the fndngs of the relatonshp between Google actvty and share returns tend to reman smlar across the days between the merger announcement and the ntal merger sgnal. Overall, n lne wth exstng lterature (e.g., Da et al., 2011), these results support the postve relatonshp between Google volume and share returns wthn the UK merger context. 10

12 3. Emprcal results 3.1 Intal fndngs Before we estmate excess returns n relaton to takeover sgnals, we offer a descrpton of our abnormal Google varables n relaton to FT coverage. Panel A of Table 3 shows the number of frms wth potental takeover actvty that was found. Followng the boxplot method, we fnd that 150 ( ) and 116 frms ( ) out of the total of 430 frms were found to be sgnaled as potental target frms. Stated dfferently, the boxplot method offers outlers only n frms that experence sgnfcant abnormal upward changes n Google volume. FT offers rumor artcles for 127 frms. Interestngly, we fnd that the frst sgnal of potental merger actvty s on average -20 days pror to the merger for both Google varables and -16 days for FT, showng that Google search volume of frms on average ncreased sgnfcantly before rumor artcles were publshed on FT. Panel B of Table 3 also explores whether Google ndcators and FT dentfy the same frms as potental target frms. For example, and FT smlarly dentfy that 249 frms are or are not to become targets, whle such ndcators show a dfferent outcome n the remanng 181 frms (a total of 430 frms). Both measures therefore ndcate the same sgnal n merely 58 percent of the frms. We therefore conclude that there s a varaton of frms that Google varables and FT dentfy as potental targets, wth sgnals beng relatvely close between and from constructon, showng that Google abnormal upward changes do not smply reflect FT s coverage of rumors. Panel C of Table 3 also explores frms for whch both Google ndcators and FT sgnal potental merger actvty, and shows whch of the sgnals appears frst. In lne wth the above results, we fnd that Google ndcators seem to precede those of FT; as an example, out of 48 11

13 frms wth and FT sgnals, frms. Results are even stronger n favor of shows a takeover sgnal frst n 34 of those n relaton to FT (22 out of 28 frms). Overall, to some extent n lne wth Da et al. (2011) and Bank et al. (2011), these results show that Google ndcators seem to capture nvestor attenton earler than meda coverage. 3.2 Estmaton of excess returns We then follow an event study analyss to explore the abnormal returns pror to the merger announcement and pror to the abnormal upward change n Google actvty. Table 4 shows the daly returns untl fve days before the merger announcement (day 0) and the cumulatve abnormal returns (CAR) every ten days over the thrty days pror to the merger. Excess returns at day t ndcate a smple arthmetc average across all target frms for that day and CAR( j, k ) ndcates the sum of those daly excess returns between day j and k. Panel A of Table 4 shows the excess returns n relaton to the merger announcement and ndcates that n lne wth the lterature (e.g., Gupta and Msra, 1989; Mathur and Waheed, 1995; Kng, 2009), share prces of target frms ncrease sgnfcantly before ther merger announcement, where the rate of ncrease s hgher closer to the merger announcement. For example, cumulatve market excess returns ( ER M ) are 5.09, 2.69, and 1.53 n the ntervals (-1,-10), (-11,-20), and (-21,-30), respectvely. CAR (-1,-30), whch are to be explaned by Google ndcators, are 9.30 ( ER M ), ( ER CAPM ), and ( ER FF 3 sgnfcant at the 1 percent level. ) percent, wth returns beng statstcally We then explore whether (Panel B of Table 4) and (Panel C of Table 4) can explan such proftablty of target frms by estmatng excess returns before the frst / sgnal (when there s one avalable) or otherwse before the merger announcement day. Stated dfferently, day 0 reflects the day wth the frst merger 12

14 sgnal, otherwse the merger announcement day. For brevty reasons we only dscuss the market excess returns ( ER M ), snce conclusons are dentcal for alternatve excess estmatons ( ER CAPM, ER FF 3 ). CAR (-1,-30) are 5.91 pror to and 7.13 pror to (versus 9.30 percent before the merger announcement), wth all returns remanng economcally and statstcally sgnfcant at the 1 percent level. Our results therefore show that Google ndcators used n the study fal to fully explan the target prce run ups pattern, showng that the ncrease of target share prces could not be predcted by market partcpants. We fnd that Google ndcators can explan at best only 36 percent of the target prce run ups at the tme nterval between -1 and -30 ( / ER M scenaro). We then explore whether Google can explan a larger percentage of excess proftablty than conventonal FT coverage, and we therefore compare excess returns between FT coverage and (Panel D of Table 4), and FT coverage and (Panel E of Table 4). Results show that Google ndcators show a sgnal a few days earler than FT, and the dfference n excess proftablty between FT and / s economcally and statstcally sgnfcant at the 1 percent level when estmatng CAR (-1,- 10). CAR (-1,-30) also show that excess returns are always lower for Google ndcators n relaton to those found for FT. Such dfferences may be economcally sgnfcant, wth returns varyng from to dependng on the excess return measure followed, but they are statstcally nsgnfcant at the 10 percent level. 10 Overall, we fnd that Google ndcators explan a larger percentage of targets excess returns before ther merger than FT. Nevertheless, Google ndcators can only capture a relatvely small part of the target prce pattern. 10 For example, the excess return of percent shown n Panel D of Table 4 has a p-value equal to

15 3.3 Robustness tests We undertake a number of tests to explore the robustness of our key results. We fnd that 6.28 percent of the tckers are nosy, such as ce, boy, and fee, and, n lne wth Da et al. (2011), we exclude them from the sample and re-estmate pror analyses. Panel A of Table 5 shows the results. Notce that due to space consderatons, we only show excess returns for CAR (-1,-10) and CAR (-1,-30) pror to,, and the dfference n correspondng returns between Google ndcators and FT. We fnd that after controllng for nosy tckers, results are smlar wth those reported n Table 4. Google ndcators may fal to explan n full the target prce run ups, but Google ndcators explan a larger percentage of such proftablty than FT. Cumulatve abnormal returns n the nterval between -1 and -10 are always economcally and statstcally lower for correspondng excess returns for FT. and than In addton, we re-estmate pror tests based only on 345 frms wth 100 percent merger actvty. Ths test explores whether pror results were drven by the sample selecton of target frms wth over 50 percent merger actvty. Panel B of Table 5 shows that pror results hold wthn such subsample. We further conduct tests of the stablty of our results durng the sample perod snce the sample ncludes the fnancal crss. Fgure 3 explores the annual cumulatve excess returns n the nterval between (-1,-10) and (-1,-30) for FT coverage,, and. We fnd that results are strong regardng CAR (-1,-10), wth excess proftablty beng lower for Google ndcators than FT n all seven years. Regardng CAR (-1,-30), results are weaker n favor of and, snce Google ndcators dsplay respectvely lower excess returns n fve and four (out of seven) years than those reported n FT. We further explore the mpact of fnancals on pror results. Fnancals may have receved ncreased 14

16 Google search volumes durng the fnancal crss and therefore, an ncrease n Google actvty n a fnancal frm may be drven due to the crss rather than to a takeover rumor. We exclude from the sample 47 fnancals and re-estmate excess returns. Panel C of Table 5 shows that results reman smlar wthn the non-fnancal subsample and we therefore conclude that the fnancal crss does not drve the results of the study. We further explore whether our results are drven by frms that reported fnancal year end results close to the merger announcement, snce such frms may face ncreased Google search actvty due to such nformaton rather than to a takeover sgnal. We exclude 64 frms wth fnancal results reported up to 60 days pror to ther merger announcement and reestmate excess returns. Panel D of Table 5 shows the results. We fnd that excess returns are reduced slghtly across portfolos, showng that Google search actvty s related wth frms fnancal year end results, however pror determned prce patterns hold strong wthn the subsample. Overall, our robustness tests support our pror conclusons on the sgnfcance of Google ndcators capturng target frms earler than FT, but wthout managng to explan the target prce run ups pattern. 4. Concluson A number of studes (e.g., Kng, 2009) have found that the share prce of target frms ncreases before ther merger announcements. Studes that have explored whether nvestors could have predcted the target frms have used meda coverage as a proxy, wth nvestors managng to predct a merger as long as a rumor was reported before the announcement. Based on the dffculty of capturng all nformaton avalable to nvestors, especally n recent years, when nternet resources and chat dscussons are heavly used, we followed an 15

17 alternatve approach. If nvestors encounter a rumor of a potental merger, most nvestors may use Google to search for further nformaton on the target company before proceedng wth a transacton; therefore, frms that feature n a rumor are expected to experence an abnormal ncrease n Google search actvty. We used the outler lterature, and more specfcally the boxplot method (Tukey, 1977), to dentfy the days that target frms experence an abnormal upward change n Google use of targets tckers, sgnalng potental merger actvty. We then followed an event study analyss to estmate excess returns before the merger announcement and before the Google takeover sgnals. We found that Google ndcators tend to sgnal target frms earler than FT; therefore, the excess returns pror to Google ndcators are lower than those reported pror to FT. Nevertheless, we found that Google ndcators cannot explan the target prce run ups and can capture at best merely 36 percent of the ncrease n the target prce pattern. Although we dd not explore nsders transactons, part of the remanng upwards prce pattern could be attrbuted to prvate nformaton. The fndngs of the study are of nterest to regulators. The Takeover Panel has been responsble for the takeover rules n the UK snce As an example, on 19 th September 2011 the Panel mplemented amendments on the exstng takeover code to mnmze the target prce run ups and amongst others, target frms were gven the responsblty to make publcally avalable any bd approach. 11 Future research can explore whether such changes to the regulatons may ncrease the sgnfcance of publc nformaton on explanng the UK target prce run ups accessed September 2012). (last 16

18 References Bank, M., Larch, M., & P. Georg. (2011). Google search volume and ts nfluence on lqudty and returns of German stocks. Fnancal Markets and Portfolo Management, 25, Chou, H., Tan, G., & X. Yn. (2010). Rumors of mergers and acqustons: Market effcency and markup prcng. Workng paper. Clarkson, P., Joyce, D., & I. Tuttcc. (2006). Market reacton to takeover rumour n nternet dscusson stes. Accountng and Fnance, 46, Da, Z., Engelberg, J., & P. Gao. (2011). In search of attenton. Journal of Fnance, 66, Da, Z., Engelberg, J., & P. Gao. (2012). In search of fundamentals. Workng Paper. Drake, M., Roulstone, D., & J. Thornock Investor nformaton demand: Evdence from Google searches around earnngs announcements. Journal of Accountng Research (forthcomng). Dyck, A., Volchkova, N., & L. Zngales. (2008). The corporate governance role of the meda: Evdence from Russa. Journal of Fnance, 63, Fama, E., & K. French. (1993). Common rsk factors n the returns on stocks and bonds. Journal of Fnancal Economcs, 33, Ferguson, N., Guo, J., Lam, H., & D. Phlp. (2011). Meda sentment and UK stock returns. Workng paper. Gupta, A., & L. Msra. (1989). Publc nformaton and pre-announcement tradng n takeover stocks. Journal of Economcs and Busness, 41, Holland, K., & L. Hodgknson. (1994). The pre-announcement share prce behavor of UK takeover targets. Journal of Busness Fnance and Accountng, 21,

19 Jensen, M., & R. Ruback. (1983). The market for corporate control: The scentfc evdence. Journal of Fnancal Economcs, 11, Keown, A., & J. Pnkerton. (1981). Merger announcements and nsder tradng actvty: An emprcal nvestgaton. Journal of Fnance, 36, Kng, M. (2009). Prebd run-ups ahead of Canadan takeovers: How bg s the problem? Fnancal Management, Wnter, Mathur, I., & A. Waheed. (1995). Stock prce reactons to securtes recommended n Busness Week s Insde Wall Street. Fnancal Revew, 30, Pound, J., & R. Zeckhauser. (1990). Clearly heard on the street; The effect of takeover rumours on stock prces. Journal of Busness, 63, Sganos, A., & M. Papa. (2012). FT coverage and target prce run-ups: Evdence from the London Stock Exchange. Journal of Busness Fnance and Accountng (2 nd Round). Tukey, J. (1977). Exploratory data analyss. Addson-Wesley. Wggns, J., & N. Hume. (2/12/2006). Premer Foods n talks over GBP1bn takeover of RHM. Fnancal Tmes. Wggns, J. (3/12/2006). RHM and Premer close to deal. Fnancal Tmes. Zvney, T., Bertn, W., & K. Torabzadeh. (1996). Overreacton to takeover speculaton. Quarterly Revew of Economcs and Fnance, 36,

20 Table 1 Descrptve statstcs of Google ndcators Average 0.03% 0.05% Medan 0.00% 0.00% Mn % % Max 69.31% 69.31% Standard devaton 8.40% 10.53% Skewness Kurtoss Kolmogorov Smrnov 0.09*** 0.11*** Notes: Ths table offers the descrptve statstcs of ndcators generated by Google Data. shows the daly change n Google volume between days -40 and 0 (day 0 = merger announcement day) and shows the abnormal daly change between days -40 and 0 above the normal Google actvty for each frm as estmated by the medan number of searches for each frm between -41 and -50 days before the merger announcement. *** shows sgnfcance at the 1 percent level. 19

21 Table 2 Google actvty and target share returns ER M CAPM ER FF 3 ER ER M CAPM Panel A: Impact of Google merger sgnal on share returns (%) ER ER FF 3 t 1.68*** 1.64*** 1.59*** 1.74*** 1.73*** 1.66*** t ** 0.94** 0.99** 1.57** 1.57** 1.40** t * 0.66* 0.58* 0.81* 1.01** 0.91* t * 1.01* 0.97* Panel B: Regresson analyss N 3,461 3,461 3,461 2,345 2,345 2,345 a ** 0.038** 0.033* a *** 0.086*** 0.085*** a 0.038** 0.039** 0.037** a ** ** ** 4 F-stat 2.082* 2.001* *** 3.988*** 4.083*** Notes: Panel A explores whether Google s abnormal upward changes n and have an mpact on share returns on the day that those sgnals were dentfed as well as over the followng three days; t + 1, t + 2, and t + 3. Excess returns ndcate a smple arthmetc average across all target frms for that day. Panel B explores whether there s n general a relatonshp between share returns and Google actvty by estmatng ER a + a AGoogle + a AGoogle + a AGoogle + a AGoogle + u t = 0 1 t 2 t 1 3 t 2 4 t 3 where all three measures are used to estmate excess returns ER t ( ER M, ER CAPM and ER FF 3 ) and both Google s volume are employed ( and ). The regresson s estmated on frms after the ntal takeover sgnal and only the slope coeffcents are presented for brevty reasons. N shows the number of observatons used. We follow the boxplot method (Tukey, 1977) to dentfy abnormal upward changes for and, where shows the daly change n Google volume and shows the abnormal daly change above the normal Google actvty for each frm as estmated by the medan number of searches for each frm between -41 and -50 days before the merger announcement. ER M shows the dfference between share and market (FTSE All Share) returns, ER CAPM shows the Captal Asset Prcng Model s rsk-adjusted returns, and ERFF 3 shows the three-factor model s rsk-adjusted returns. *, **, and *** show sgnfcance at the 10, 5, and 1 percent levels. t 20

22 Table 3 Comparson of takeover sgnals FT cov erage Panel A: Coverage of frms Wth a takeover sgnal Wthout a takeover sgnal Average days Medan days Panel B: Agree/dsagree sgnal of merger actvty 310/ / /187 Panel C: Frst sgnal of a takeover actvty vs FT cov erage vs FT cov erage 22 6 vs Notes: Ths table compares sgnals of takeover actvty among,, and FT cov erage. Panel A shows the number of frms wth and wthout a takeover sgnal and how many days before the merger (day 0) those sgnals occur. Panel B explores the extent to whch alternatve ndcators predct the same outcome as to whether a frm would become a target frm. As an example and FT smlarly dentfy the outcome n 249 frms, whle the outcome dffers at the remanng 181 frms. Panel C explores whch of the sgnals appears frst. shows the daly change n Google volume and shows the abnormal daly change above the normal Google actvty for each frm as estmated by the medan number of searches for each frm between -41 and -50 days before the merger announcement. 21

23 Table 4 Estmaton of excess returns (%) ER M CAPM ER FF 3 ER ER M CAPM ER ER FF 3 Panel A: In relaton to merger announcement Panel B: In relaton to *** 15.86*** 15.80*** 10.25*** 10.37*** 10.32*** *** 2.30*** 2.23*** 1.34*** 1.32*** 1.27*** *** 0.88*** 0.94*** 0.43** 0.50*** 0.58*** * 0.32** * 0.42** 0.29* *** 0.64*** 0.60*** CAR (-1, -10) 5.09*** 5.56*** 5.25*** 2.43*** 3.03*** 2.68*** CAR (-11, -20) 2.69*** 3.05*** 2.75*** 2.70*** 3.20*** 2.97*** CAR (-21, -30) 1.53*** 2.21*** 2.04*** *** 1.54*** CAR (-1, -30) 9.30*** 10.82*** 10.04*** 5.91*** 7.88*** 7.20*** Panel C: In relaton to Panel D: FT *** 11.97*** 11.94*** -2.50* -2.48* -2.57* *** 1.53*** 1.52*** -3.44*** -3.46*** -3.34*** *** 0.52*** 0.63*** -0.52** -0.55** ** 0.50*** 0.39** ** 0.29* CAR (-1, -10) 3.60*** 4.08*** 3.87*** -4.02*** -4.07*** -3.97*** CAR (-11, -20) 2.28*** 2.64*** 2.35*** * 1.66* CAR (-21, -30) 1.25*** 2.02*** 1.92*** CAR (-1, -30) 7.13*** 8.74*** 8.14*** Panel E: FT *** -3.25*** -3.10*** * CAR (-1, -10) -2.86*** -3.02*** -2.78*** CAR (-11, -20) CAR (-21, -30) CAR (-1, -30) Notes: Ths table shows the estmaton of excess returns pror to the merger announcement (Panel A), pror to (Panel B), and pror to (Panel C). Therefore 0 represents the merger announcement day at Panel A and ether the day wth the frst / sgnal or otherwse the merger announcement day at Panels B and C. -1, -2, -3, -4 and -5 show days pror day 0. Panels D and E compare the excess returns found n FT n comparson to those reported n and, respectvely. shows the daly change n Google volume and shows the abnormal daly change above the normal Google actvty for each frm as estmated by the medan number of searches for each frm between -41 and -50 days before the merger announcement. Excess returns at day t ndcate a smple arthmetc average across all target frms for that day and CAR ( j, k) ndcates the sum of those daly excess returns between day j and k. We follow alternatve rsk-adjusted excess returns: ER shows the dfference M 22

24 between share and market (FTSE All Share) returns, ERCAPM shows the Captal Asset Prcng Model s rskadjusted returns, and ERFF 3 shows the three-factor model s rsk-adjusted returns. *, **, and *** show sgnfcance at the 10, 5, and 1 percent levels. 23

25 In relaton to In relaton to Table 5 Robustness tests (%) ER M CAPM ER FF 3 Panel A: Excludng nosy tckers ER ER M CAPM ER FF 3 Panel B: Excludng frms wth less than 100% merger actvty ER ER M CAPM CAR (-1,-10) ER FF 3 Panel C: Excludng fnancals ER ER M CAPM ER ER FF 3 Panel D: Excludng frms wth fnancal year end results 60 days before ther merger announcement 2.53*** 3.22*** 2.82*** 3.06*** 3.69*** 3.31*** 2.89*** 3.43*** 2.91*** 2.33*** 2.87*** 2.40*** 3.78*** 4.35*** 4.11*** 3.92*** 4.56*** 4.26*** 4.09*** 4.49*** 4.11*** 3.07*** 3.43*** 3.18*** FT -4.12*** -4.14*** -4.09*** -3.90*** -4.01*** -3.88*** -4.17*** -4.36*** -4.40*** FT -2.87*** -3.01*** -2.80** -3.04*** -3.14*** -2.93*** -2.97*** -3.30*** -3.20*** In relaton to In relaton to CAR (-1,-30) -3.79*** -3.81*** -3.75*** -3.05*** -3.25*** -2.97*** 6.03*** 8.23*** 7.35*** 6.94*** 9.16*** 8.43*** 6.58*** 8.28*** 7.27*** 5.76*** 7.23*** 6.36*** 7.33*** 9.15*** 8.37*** 8.41*** 10.48*** 9.85*** 7.83*** 9.13*** 8.19*** 6.56*** 7.61*** 6.90*** FT FT Notes: Ths table shows the excess returns of three robustness tests when excludng nosy tckers (Panel A), when excludng target frms wth less than 100 percent merger actvty (Panel B), when excludng fnancals (Panel C) and when excludng frms wth fnancal year end results up to 60 days pror to ther merger announcement (Panel D). CAR (j,k) ndcates the cumulatve excess returns n the nterval between days j and k. shows the daly change n Google volume and shows the abnormal daly change above the normal Google actvty for each frm as estmated by the medan number of searches for each frm between -41 and -50 days before the merger announcement. We follow alternatve rsk-adjusted excess returns: ERM shows the dfference between share and market (FTSE All Share) returns, ERCAPM shows the Captal Asset Prcng Model s rsk-adjusted returns, and ERFF 3 shows the three-factor model s rsk-adjusted returns. ** and *** show sgnfcance at the 5 and 1 percent levels. 24

26 Fg. 1. Google trend search for RHM plc Notes: Ths fgure represents the daly output for a Google trend search of RHM between October and December Hgh/low hstorc trend ndcates hgh/low numbers of searches n Google. Notce that RHM plc was acqured by Premer Foods plc on 4 th December

27 Fg. 2. Cumulatve abnormal Google searches Cumulatve abnormal Google searches (%) Cumulatve abnormal Google searches (%) Days Days Notes: Ths fgure shows the cumulatve Google volume untl the merger announcement day (day 0). shows the daly change n Google volume and shows the abnormal daly change above the normal Google actvty for each frm as estmated by the medan number of searches for each frm between -41 and -50 days before the merger announcement. 26

28 Fg. 3. Annual excess returns ( ER FF 3 ) CAR(-1,-10) CAR(-1,-30) Cumulatve returns (%) FT Cumulatve returns (%) Google2 Google1 FT 0.00 Google Google Years Years Notes: Ths fgure shows the annual cumulatve excess returns n the nterval between (-1,-10) and (-1,-30) for FT coverage, sample perod. shows the daly change n Google volume and, and durng the shows the abnormal daly change above the normal Google actvty for each frm as estmated by the medan number of searches for each frm between -41 and -50 days before the merger announcement. For brevty reasons, results are shown only for ER, whch shows the three-factor model s rsk-adjusted returns (conclusons reman unchanged when alternatve excess return methods are estmated). FF 3 27