How Important Is the Financial Media in Global...
Transcript of How Important Is the Financial Media in Global...
How Important Is the Financial Media inGlobal Markets?
John M. GriffinUniversity of Texas at Austin, McCombs School of Business
Nicholas H. HirscheyUniversity of Texas at Austin, McCombs School of Business
Patrick J. KellyNew Economic School and University of South Florida
Thisarticle studies differences in the information content of 870,000 news announcementsin 56 markets around the world. In most developed markets, a firm’s stock price movesmuch more on days with public news about the firm. In contrast, in many emerging marketsvolatility is similar on news and non-news days. We examine several hypotheses for ourfindings. Cross-country differences in stock price reactions are best explained by insidertrading, followed by differences in the quality of the news dissemination mechanism. Ourfindings are useful for quantifying the extent of insider trading and how the financial mediaaffects international markets. (JEL G14, G15)
Public news announcements are a major mechanism for disseminating in-formation to investors. Each day the financial media releases thousands ofarticles covering companies in markets worldwide. Investors use this news toestimate assets’ fundamental values. Despite the perceived importance of the
The first two authors are from the McCombs School of Business at the University of Texas at Austin.Kelly is at the New Economic School in Moscow and the University of South Florida in Tampa. Wethank Utpal Bhattacharya, Bernie Black, Zhiwu Chen, Craig Doidge, Alexander Dyck, Miguel Ferreira,Gary Gebhardt, Jay Hartzell, Ravi Jagannathan, Yu Jing, Lidia Kelly, Alok Kumar, David Lesmond, MarcLipson, Christian Leuz, Xi Liu, Lilian Ng, Peter Mackay, Federico Nardari, David Ng, Clemens Sialm, StefanSiegel, Paul Tetlock, Sheridan Titman, and seminar participants at the Chinese University of Hong Kong,Chulalongkorn Accounting and Finance Symposium, Darden International Conference, International FinanceConference at McGill University, Fudan University, Jiao Tong University, Hong Kong University of Scienceand Technology, Nanyang Technological University, New Economic School, Peking University, SingaporeManagement University, University of Melbourne, University of Texas at Austin, and University of Washingtonfor helpful comments and discussion. We are grateful to Mikael Bergbrant, Garret Fair, and Jordan Nickersonfor their diligent research assistance. Send correspondence to John M. Griffin, University of Texas at Austin,McCombs School of Business, 1 University Station B6600 Austin, TX 78746; telephone: (512) 471-6621; fax:(512) 471-3034. E-mail: [email protected].
c© The Author 2011. Published by Oxford University Press on behalf of The Society for Financial Studies.All rights reserved. For Permissions, please e-mail: [email protected]:10.1093/rfs/hhr099 Advance Access publication October 25, 2011
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
financialmedia, there has been little attempt to either quantify its importanceinternationally or to understand why its impact may vary across countries.
This article studies differences in the information content of public newsannouncements in international equity markets. We quantify how the marketreaction to these announcements varies across countries and examine potentialexplanations for these differences. These analyses deepen our understanding ofdifferences in the international information environment and how informationis disseminated to investors.
There are several reasons why the market reaction to news announcementsmay differ across countries. First, there may be differences in how wellinvestors anticipate the information in a news announcement using publicnews channels, such as peer firms’ news announcements. Second, the pricemay incorporate the information in a news release prior to its announcementbecause of insider trading. Third, if journalists are more sophisticated in certaincountries, one might expect news coverage in countries with sophisticatedjournalists to provide a more precise signal and lead to larger market reactions.Fourth, accounting quality may differ across countries. We find the mostsupport for insider trading and, to a lesser extent, differences in the qualityof the news dissemination mechanism.
We start by collecting a large sample of articles about international stocksfrom the Factiva news archive. Our sample of general news articles fromJanuary 2003 to June 2009 covers 2,593 firms, with 572,987 news articles in26 developed markets and 298,614 in 30 emerging markets.
To facilitate comparison across stocks, we first study a news event that iscommon to all firms—earnings announcements. We ensure the accuracy ofBloomberg earnings announcement dates through Factiva by a simple algo-rithm, which leads to extremely accurate earnings announcement dates. Wefind that volatility on event days varies substantially both between developedand emerging markets and within each category. Stock price moves range from50% more than normal volatility in a number of developed markets (Denmark,the United Kingdom, Sweden, the Netherlands, the United States, Finland,Hong Kong, and Germany) to less than 5% more than normal in severalemerging markets (Thailand, Turkey, Mexico, and Indonesia). China exhibitsearnings reactions only 12% above normal.
Second, we extend our study to all news articles. News is generally moreuseful for explaining idiosyncratic stock return variation in developed marketsthan in emerging markets. We find that our results are not driven by earningsannouncements, time-stamp conventions, or the type of international financialnews.
Third, we use cross-country variables related to the four main hypothe-ses regarding public informed trading, insider trading, news dissemination,and accounting quality. We investigate cross-country regressions with bothearnings and general news. With the stock price reactions to earnings newsand standard cross-country regressions, we find that a survey measure of the
3942
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
prevalence of insider trading, a measure of technological development thatwe interpret as relating to the speed of news dissemination, and accountingstandards are most pertinent. We then investigate a wider set of variablesrelating to our main hypotheses and other variables commonly used in cross-country studies through the Bayesian Stochastic Search Variable Selection(SSVS) methodology ofGeorge and McCulloch(1993). The SSVS selectsimportant explanatory variables based on a combination of their economicand statistical significance while incorporating skepticism to account for theprobability a variable is picked by chance. For earnings news, the procedureconfirms that insider trading, technological development, and accountingstandards are highly important. When investigating the importance of all typesof news, accounting standards are no longer significant and insider tradingand technological development stand out as the most important determinants.Overall, the cross-country regressions show the most support for the insidertrading hypothesis, followed by technological development as a proxy for thenews dissemination mechanism.
We investigate two other hypotheses related to insider trading to furtherscrutinize our findings. If insider trading is part of the reason that public newsevents are not as important in low-news-reaction countries, then we should seelarger price run-ups ahead of mergers as insiders trade on their knowledge ofthe impending acquisition. Consistent with this prediction, markets in whichstock prices have little response to earnings news have nearly three timesmore pre-announcement news leakage than markets where earnings news isimportant. Second, in countries where non-public information channels aremore important, we expect relatively more informed trading on days withoutpublic news. In countries with high average earnings announcement reactions(where prices move more with public news events), we find evidence thatreversals concentrate around non-news days, similar to prior studies using onlyU.S. data. However, in markets where prices respond little to news, we findthat non-news days have smaller reversals. This evidence is consistent withrelatively more informed trading and proportionally less liquidity trading inlow-reaction countries than in high-reaction countries.
This study is related to a number of studies that examine the role of themedia in financial markets.Roll (1988) shows that stocks exhibit similarlylow market-modelR2 on days with and without news and finds that residualvolatility is only slightly higher on news days.Chan(2003) andTetlock(2010)find that stock returns reverse only when the initial price move occurs whenthere is no news about the stock.Tetlock(2007) andTetlock, Saar-Tsechansky,and Macskassy(2008) find that the market rapidly incorporates most of theinformation associated with the linguistic content of news articles.1Huberman
1 Busheeet al. (2010) find that the initiation of media coverage leads to more trading and larger absolute stockreturn movements for NASDAQ stocks around earnings announcements.Bhattacharya et al.(2009) show thatdifferences in media coverage did not drive differences in Internet stock valuations during the NASDAQ bubble.Fang and Peress(2009) find that stocks with no news earn higher abnormal returns.
3943
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
andRegev(2001) andTetlock (2011) provide evidence that even stale newsmoves prices. A common theme in this U.S.-based literature is that news, evenstale news, can drive stock price movements.
Much less is known about news for stocks in international markets.Bhattacharya et al.(2000) examine 75 hand-collected news announcementdates in Mexico from 1994 to 1997 and find no stock price reaction in the newsannouncement window. In addition to a number of individual country studiesof earnings announcements,DeFond, Hung, and Trezevant(2007) examineearnings reaction differences across 26 primarily developed markets. However,their test design, focus, and methodologies are quite distinct from those weemploy.2 Bailey, Karolyi, and Salva(2006) look at earnings event reactionsbefore and after a U.S. cross-listing and find that U.S. accounting requirementslead to increases in earnings announcement reactions.
Our article also fits into a large literature that seeks to understand howemerging and developed markets differ in terms of various aspects of effi-ciency. Despite a large international literature, there is relatively little under-standing of differences in the informational environment across countries, andfew studies that attempt to quantify the importance of public news interna-tionally.3,4 Griffin, Kelly, and Nardari(2010) find that it is not feasible tomake statements of relative market efficiency internationally using traditionalmeasures unless one can control for the information environment. Our findingsquantify one crucial aspect of the information environment and suggest that themagnitude of news response can be used to identify the presence of informedtrading.
Our article also relates to a growing literature that seeks to documenttrading on inside information in the United States.Griffin, Shu, and Topaloglu(2011) suggest that evidence of trading on information leaked from investmentbankers to their clients prior to takeovers and earnings announcements in theUnited States is less prevalent than is suggested by prior academic work andthe U.S. financial media. Our findings suggest that insider trading prior totakeovers is much more prevalent in emerging and some smaller developedmarkets, where stocks respond little to earnings news.
2 Thereare three important differences with our study. First, we find that I/B/E/S earnings announcements dates,such as those used byDeFond, Hung, and Trezevant(2007), are on average accurate only 23% of the time indeveloped non-U.S. markets. Second, they examine stock price reactions in 23 developed markets with onlythree emerging markets. Third, they examine earnings reactions only, whereas we also focus on general newsevents, takeovers, and return reversals.
3 Using international microstructure data,Lia, Ng, and Zhang(2009) find small differences in the probability ofinformed trading between developed and emerging markets.
4 Gagnonand Karolyi(2009) examine the relation between reversals and continuation (associated with liquidityand informed trading) and trading volume around the world.Lang, Lins, and Miller(2003) find that cross-listingimproves a firm’s information environment as proxied for by analyst coverage and accuracy, andBae, Bailey,and Mao(2006) find that analyst coverage increases following liberalization. The enforcement of insider tradinglaws has been linked to a lower cost of capital (Bhattacharya and Daouk 2002) and higher idiosyncratic volatilityin developed markets (Fernandes and Ferreira 2009).
3944
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
Section1 develops our hypotheses. Section2 describes the data sources.Section3 displays our event reactions for earnings events, and Section4reports results for general news. In Section5, we test our hypotheses withcross-country data, while Section6 provides additional tests related to insidertrading. Section7 concludes.
1. Hypothesis Development
We test four hypotheses for why the price response to public news announce-ments might differ across countries. We believe that differences in marketreactions across countries may be driven by variation in 1) pre-announcementpublic news dissemination; 2) insider trading; 3) the quality of the newstransmission mechanism; and 4) accounting quality.
1.1 Public information disseminationPublic news channels provide substantial information about changes in firmvalue. An important channel is cross-firm learning through peer firms’ earningsannouncements (Foster 1981; Freeman and Tse 1992). Givoly and Palmon(1982) find that late-reporting firms have dampened stock price reactions,because investors learn about their earnings from earlier-reporting firms’earnings announcements.Hou (2007) documents important lead-lag patternsin intra-industry returns, andPatton and Verardo(2010) show that the order ofearnings announcements affects firm betas. In an international sample of firms,if early- or late-reporting firms concentrate in certain countries, then this mayaffect those countries’ average market reaction to earnings news.
Hypothesis H1:Countries with firms reporting late in the earnings report-ing cycle will have lower stock price responses to news.
To capture the importance of the earnings reporting cycle, we compute eachfirm’s reporting order relative to all other firms in the global industry. We usethis reporting-order variable in firm-level regressions and the country averagein country-level regressions.
1.2 Information dissemination through insider tradingIf insiders trade on private information, prices may incorporate value-relevantinformation prior to its public release. This results in a smaller market responseto official public announcements of firm information.Bhattacharya et al.(2000), for example, conclude that insider trading is the reason the marketdoes not react to news in Mexico. On the other hand, by the start of our sampleperiod, nearly all countries in our sample have enforced insider trading laws,so there may be little variation in insider trading across countries.
Hypothesis H2A: In countries where insider trading is more prevalent,stock prices move less on announcement dates.
3945
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
We use the response from a survey of executives in each country around theworld to a question asking whether or not insider trading is common in theircountry as our primary measure for the prevalence of insider trading in cross-country analysis.
We consider additional testable hypotheses for insider trading. One testablehypothesis relates to a major standardized event, takeover announcements, thatare difficult to predict with public news. Markets with insider trading shouldhave more stock price run-up prior to the release of any public takeover news.5
Hypothesis H2B: In countries where insider trading is more prevalent,stock prices will increase more rapidly prior to the first takeover announcementdate.
Another implication is that on non-news days, prices will exhibit less reversalin markets where there is greater private informed trading.Chan(2003) andTetlock(2010) have shown that in the United States, reversals are concentratedon non-news days. For markets with relatively low levels of insider trading,such as the United States, we expect non-news-day trading to be predominatelyliquidity trading, so there should be reversals on non-news days. However,we expect there to be relatively more informed trading on non-news days inmarkets with high levels of private informed trading, which should result insmaller non-news-day reversals.
Hypothesis H2C:If cross-country differences in event reactions are relatedto informed trading, we expect little difference in reversals between news andnon-news days in markets with low event reactions.
We test these predictions both by examining reversals following sorts onextreme return events and for all return moves through cross-sectionalregressions.
1.3 News transmissionThe quality of news gathering and transmission may differ across countries.More sophisticated journalists may be better at focusing on value-relevantnews. In contrast, if most news is simply a transmission of company pressreleases with little filtering or the quality of journalists around the world issimilar, stock price reactions to news may not be affected by differences in thequality of the journalism.
Hypothesis H3A: Prices respond to news more in countries with higher-quality journalism.
5 Bris (2005)examines general patterns of stock price run-ups prior to takeovers before and after the implemen-tation of insider trading laws.
3946
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
Direct measures of journalism quality are difficult to observe, but we inves-tigate several alternatives. First, as a firm-level proxy, we use an indicator ofwhether the news was from a reputable international business news source:TheWall Street Journal, Financial Times, Dow Jones Newswires, and Reuters.6
Second,in markets with more sophisticated journalists, journalists may focustheir attention on a firm when particularly value-relevant information is beingreleased. Hence, as a proxy for the financial sophistication of the news, weuse the ratio of the number of articles in the earnings-event window relativeto the number of articles in the pre-event window.7 Third, as another proxyfor journalist sophistication, we count the number of articles with more than500 words per firm mentioned as a measure of in-depth coverage. Fourth, itis possible that in countries where there is greater press freedom, it is saferfor journalists to root out information and report it accurately. We examinecountry-level press freedom as a general proxy for journalism quality.
A related hypothesis is the means of news dissemination. News is likelydistributed faster and to a broader investor base through Internet sourcesin countries that are more technologically sophisticated, with the result thatinformation is more rapidly incorporated into prices.
Hypothesis H3B: News has a larger impact on prices in countries withgreater technological development.
1.4 Accounting qualityPrices will be more responsive to news when the content of the news hasgreater economic meaning. For earnings-related news, valuation signals areclearer when there are stronger accounting standards and better financialdisclosure because the income statements are more trustworthy. This leads toour final hypothesis:
Hypothesis H4:In countries with higher accounting quality, prices respondmore to earnings news.
We use two measures to measure accounting quality. The first is the percentageof firms that follow U.S. Generally Accepted Accounting Principles (GAAP)or International Financial Reporting Standards (IFRS). Because accountingstandards can be more stringently or less stringently applied even amongcountries with firms following IFRS, we also use a survey-based measure fromthe Global Competitiveness Report, “Strength of Accounting Standards.”
6 Dyck, Volchkova, and Zingales(2008) find that the foreign media can play a disciplining role in Russia becausefirms care about their reputation abroad.
7 We thank an anonymous referee for these helpful ideas. This variable also might be thought of more broadly asa measure of news importance.
3947
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
We examine these hypotheses first in cross-country regressions, but we alsouse firm-level data and regressions to check our inferences with firm-levelcontrols.
2. Data
The main data in this article consist of firm stock returns, news articles,earnings and takeover announcements, and country-level descriptive variables.
2.1 Preliminaries: Sample and returnsDaily returns, accounting for dividends and capital structure changes, andmarket capitalizations are from CRSP for the United States and from ThomsonFinancial’s Datastream for the rest of the world. Since we wish to restrict oursample to common equity, for the United States we use stocks with a CRSPshare code of 10 or 11. For non-U.S. securities, we follow the substantialscreens ofGriffin, Kelly, and Nardari (2010), which eliminate preferredstock, warrants, unit or investment trusts, duplicates, GDRs or cross-listings,and other non-common equity. Because we desire to capture the effects ofinformation release on stock prices, we require stocks to be actively traded;all stocks are required to have price changes on at least 50% of the tradingdays in the prior year. This way we avoid capturing returns that largely reflectstale information. This measure of liquidity, the percentage of zero returndays in a year, is the main measure of liquidity used byBekaert, Harvey, andLundblad(2007) and is similar to theLesmond et al.(1999) andLesmond(2005) transaction costs measure, but less subject to estimation problems.
2.2 News articlesWe collect a sample of news articles from the Dow Jones Factiva news archive.Factiva contains articles from over 28,000 sources published in 159 countriesand 23 languages.8 Thedatabase includes major business news publishers suchas The Wall Street Journal, Financial Times, Dow Jones, and Reuters. Weidentify news for a firm using full-text searches for the firm’s name in theheadline or lead paragraph of articles. The computational burden of the webretrieval process is quite intense, so we collect news for a random sample of5,875 firms over three windows from January 2003 to June 2009.9 Beforenewsand liquidity filters, we initially structure the developed market sample to haveapproximately 50 firms from all countries other than the United States.
8 http://www.factiva.com/sources.asp.For our dataset the news is from over 5,000 local and foreign sources.
9 First,we downloaded news from January 2003 through August 2008 for 4,365 firms from emerging markets andnon-OECD developed markets. We over-sample emerging markets, because emerging market firms receive lessmedia coverage and we want to ensure that we have a large sample with earnings announcements covered by themedia. Second, because U.S. firms represent such a large fraction of firms in the world, we downloaded newsfrom January 2004 through December 2008 for 500 U.S. firms. Third, we downloaded news from January 2004through June 2009 for 1,173 firms in countries on the MSCI’s list of developed markets in 2001.
3948
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
Table1 reports summary statistics for this sample of news. For a firm yearto be included, in addition to the 50% price change filter discussed above, werequire that a firm has at least one news article during a given year. However,to ensure a sufficient number of non-news days, no more than 75% of thetrading days during the year can have news articles. The sample consists of2,593 firms with 1,342 firms from 26 developed markets and 1,251 from 30emerging markets. This results in 572,987 news articles for developed marketstocks and 298,614 articles for emerging market stocks from 4,202 sources.This compares favorably to other firm-level news samples such asTetlock,Saar-Tsechansky, and Macskassy(2008), whose sample contains over 350,000articles about S&P 500 firms from the Dow Jones News Service andThe WallStreet Journal(also collected from Factiva). We have an average of 427 articlesper firm in developed markets and 239 in emerging markets, which translatesto at least one news article appearing for each firm on 15.8% of trading daysin developed markets and 12.2% of days in emerging markets.
2.3 Earnings announcement datesWithout accurate earnings announcement event dates, any analysis of marketreactions is flawed. For this reason, we spend considerable effort investigatingmethods for obtaining accurate earnings announcement dates.
We use an automated event checking procedure. We start with Bloombergannouncement dates because we find they are more than twice as accurateas I/B/E/S dates.10 The procedure then isolates a subset of events for whichthe date corresponds to the first release of earnings news in Factiva. Theprocedure isolates a subset of events for which the date from Bloombergcorresponds to the first release of earnings news in Factiva. We verify that eachBloomberg event date corresponds to an article announcing earnings in Factivaand exclude events where earnings news is released prior to the Bloombergannouncement date. Because of Factiva download constraints and the newsscreening restrictions, the number of events that pass our Factiva date-checkingprocedure is considerably smaller than the total Bloomberg announcements ineach country shown in the Online Appendix Table IA.2.11 Our sample rangesfrom January 2004 to September 2008 and is obviously tilted toward firms withmore financial media coverage. The automated procedure leads to earnings
10 Throughoutthe article, we refer to results that one might refer to as “in unreported results” by referring to theirposition in an Online Appendix. Viewing these tables and figures is not required for reading the article but theyare posted for completeness. In Online Appendix Table IA.3, we find that I/B/E/S dates have a confirming articlewithin a [−1, 1] day window around the date listed by I/B/E/S only 23% of the time in developed markets and8.4% in emerging markets.
11 In addition, differences in news checking accuracy across countries create differences in the percentage of eventsin a country that pass our filters. For example, as shown in Online Appendix Table IA.2, China starts out withmore firms than Malaysia, but Malaysia ends up with more events in the final sample because a higher percentageof the Malaysian events pass the accuracy filters. We aggregate most of our data to the country level to put equalweight on inferences in each market.
3949
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
T abl
e1
New
ssu
mm
ary s
tatis
tics
De v
elop
edTo
talfi
rms
Tota
lart
icle
sA
vg.%
Eng
.A
vg.%
Bus
.A
vg.%
WC
<25
0E
mer
ging
Tota
lfirm
sTo
tala
rtic
les
Avg
.%E
ng.
Avg
.%B
us.
Avg
.%W
C<
250
Aus
tral
ia20
6,15
598
.02.
657
.3A
rgen
tina
5121
,60
525
.211
.967
.1A
ustr
ia20
20,40
628
.141
.453
.9B
razi
l19
3,34
176
.539
.972
.8B
elgi
um19
13,16
320
.811
.859
.2B
ulg a
ria8
2,55
298
.21.
740
.4C
anad
a8
2,17
899
.77.
142
.2C
hile
2510
,55
451
.215
.469
.4C
ypru
s17
2,08
694
.015
.576
.9C
hina
231
31,25
596
.93.
736
.1D
enm
ark
2613
,19
828
.120
.659
.5C
olom
bia
209,
444
35.4
28.3
69.4
Fin
land
2514
,06
454
.916
.166
.0C
roat
ia2
765
82.8
21.0
59.5
Fra
nce
134,
685
28.3
15.0
64.9
Cze
chR
epub
lic9
9,39
560
.826
.369
.2G
erm
any
1315
,87
849
.012
.042
.9E
gypt
302,
795
96.3
14.5
81.0
Gre
ece
435,
496
92.8
13.2
45.0
Est
onia
131,
376
64.2
0.8
65.1
Hon
gK
ong
188
22,94
492
.919
.637
.7H
unga
ry28
12,44
266
.414
.465
.9Ir
elan
d8
4,57
396
.426
.145
.8In
dia
138
57,40
499
.57.
139
.7Is
rael
364,
684
92.8
8.4
51.2
Indo
nesi
a28
4,94
996
.512
.262
.2Ita
ly27
34,86
419
.916
.559
.1K
enya
255,
633
96.2
5.2
25.9
Japa
n39
7,93
286
.856
.179
.4La
tvia
83,
847
62.2
1.2
42.8
Net
herla
nds
2021
,70
224
.740
.765
.5Li
thua
nia
3113
,10
552
.61.
158
.0N
ewZ
eala
nd23
13,11
999
.68.
649
.1M
alay
sia
154
21,20
293
.613
.561
.9N
orw
ay23
12,16
345
.439
.265
.7M
e xic
o19
8,06
742
.329
.358
.3P
ortu
gal
1312
,94
026
.630
.660
.4M
oroc
co21
1,27
457
.226
.273
.4S
inga
pore
114
23,66
993
.016
.434
.3Pa
kist
an73
7,64
698
.27.
462
.6S
pain
2553
,05
221
.615
.471
.8P
eru
224,
010
47.9
23.0
68.9
Sw
eden
3016
,66
735
.125
.461
.5P
hilip
pine
s29
6,24
698
.418
.433
.6S
witz
erla
nd17
13,22
511
.211
.759
.8P
olan
d48
17,96
333
.15.
055
.1Ta
iwan
217
46,70
093
.649
.860
.1R
oman
ia30
1,60
091
.54.
365
.9U
.K.
173,
695
97.8
26.9
47.9
Slo
vaki
a2
1,03
267
.22.
064
.2U
.S.
341
183,
749
96.1
20.7
43.5
Slo
veni
a11
1,35
488
.57.
669
.8S
outh
Afr
ica
4611
,89
896
.912
.948
.9T
haila
nd67
11,76
097
.65.
734
.7T
urke
y58
10,82
253
.114
.179
.1V
enez
uela
53,
278
28.5
46.2
71.5
Tota
l1,
342
572,
987
1,25
129
8,61
4M
ean
51.6
22,0
38.0
62.6
21.8
56.2
41.7
9,95
3.8
71.8
14.0
59.1
Dev
.-E
mg.
−9.
27.
8−
2.9
(t-s
tat.)
(−1.
14)
(2.2
8)(−
0.83
)
We
iden
tify
firm
new
sus
ing
full-
text
sear
ches
for
the
firm
’sna
me
inth
ehe
adlin
eor
lead
para
grap
hof
artic
les
inF
activ
a.F
irmna
mes
are
the
Dat
astr
eam
expa
nded
nam
efie
ldw
ithco
mm
onch
arac
ter
strin
gs(A
,DE
AD
,CO
,SA
,etc
.)re
mov
ed.T
hesa
mpl
eis
cons
truc
ted
asdi
scus
sed
inS
ectio
n2
and
foot
note
9.“A
vg.%
Eng
.”is
the
aver
age
perc
enta
geof
artic
les
ina
firm
year
writ
ten
inE
nglis
h.“A
vg.%
Bus
.”is
the
aver
age
perc
ento
fart
icle
sin
afir
mye
arth
atar
epu
blis
hed
byD
owJo
nes,
Reu
ters
,the
Fin
an
cia
lTim
es,
orT
he
Wa
llS
tre
etJ
ou
rna
l.“
Avg
.%W
C<
250”
isth
eav
erag
epe
rcen
tofa
rtic
les
ina
firm
year
with
less
than
250
wor
ds.A
coun
try
iscl
assi
fied
asde
velo
ped
ifth
egr
oss
natio
nali
ncom
epe
rca
pita
isgr
eate
rth
an10
,725
U.S
.dol
lars
inca
lend
arye
ar20
05,f
ollo
win
gth
eW
orld
Ban
k’s
clas
sific
atio
nfo
r“h
igh
inco
me”
coun
trie
s.A
llot
her
coun
trie
sar
ecl
assi
fied
asem
ergi
ng.O
EC
Dm
embe
rshi
pis
from
cale
ndar
year
2005
.
3950
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
dateswith considerable accuracy (as discussed in Online Appendix A andTable IA.5).
Panel A of Table2 shows the number of events by country. Our final sampleof earnings announcements contains 3,504 events from 26 developed marketsand 2,079 events from 30 emerging markets. Despite the focus on accuracy, thesample seems to have broad coverage across markets. Panel B of Table2 showsthe number of articles around earnings announcements for each of three sizebins. Each December we sort all U.S. stocks listed on NYSE/AMEX/NASDAQinto equal portfolios based on dollar-denominated size breakpoints. In thesmall- and large-size bins, developed markets have slightly more news articlesin the non-earnings announcement window. However, in the announcementwindow the news discrepancy is large, with developed markets exhibitingroughly one more news article per day for small and medium firms, and over2.6 more articles per day in the large-size bin.
2.4 Takeover announcement datesThe sample of initial merger announcements is composed of the earliest datefrom among three sources: SDC, FactSet, and Bloomberg.12 Additionally, wecheck our dates by making sure there is no article in Factiva with both thefirm’s name and a merger keyword in the headline or lead paragraph from 60to two calendar days before the merger. For the 22 languages we check articlesin, we end up with over 1,500 translated merger words, as discussed in OnlineAppendix IA.2. Panel A of Table2 shows that we obtain 466 merger eventsfrom 23 developed markets and 105 events from 11 emerging markets.
2.5 Country-level variablesWe use country-level variables from many sources that have been used in theprior international literature. We describe these variables in detail in AppendixTable A1. Major sources include the World Bank Development Database andthe World Economic Forum’s Global Competitiveness Report (GCR). TheGCR reports the results of a survey of over 4,000 executives in 59 countries.When available, we average annual levels over 2003 through 2008 to reduceyear-to-year noise in these variables.13
3. Earnings Announcements and Stock Returns across Countries
We wish to quantify the relevance of information disseminated on newsdays. Earnings announcements are advantageous in that they are a common
12 We require targets to have at least one article in the news from 60 calendar days to two trading days before theevent to ensure that the lack of an article discussing rumored mergers is not due to errors that prevented us fromdownloading any news for the firm.
13 Onenotable exception is a GCR survey question that asks executives if “insider trading in your country’s stockmarkets is (1=pervasive, 7=extremely rare).” We average the responses in the editions from 1999, 2000, and2002–2003, because the question was not surveyed in later years.
3951
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
Tabl
e2
Ear
ning
san
dm
erge
ran
noun
cem
ents
sum
mar
yst
atis
tics
Pan
elA
:Firm
and
Eve
ntCou
nts
Dev
elop
edC
ount
ries
Ear
ning
sF
irms
Ear
ning
sE
vent
sM
erge
rE
vent
sE
mer
ging
Cou
ntrie
sE
arni
ngs
Firm
sE
arni
ngs
Eve
nts
Mer
ger
Ev
ents
Aus
tral
ia25
3958
Arg
entin
a29
66−
Aus
tria
2648
−B
razi
l20
34−
Bel
gium
4810
03
Chi
le30
63−
Can
ada
3460
125
Chi
na16
727
0−
Den
mar
k45
838
Col
ombi
a4
4−
Fin
land
5710
64
Cro
atia
13
−F
ranc
e44
7812
Cze
chR
epub
lic1
1−
Ger
man
y21
3724
Egy
pt15
182
Gre
ece
3350
17E
ston
ia4
4−
Hon
gK
ong
319
516
37H
unga
ry19
37−
Irel
and
2344
1In
dia
224
307
33Is
rael
3448
2In
done
sia
4476
1Ita
ly64
129
12K
enya
89
−Ja
pan
105
250
−La
tvia
23
−N
ethe
rland
s65
130
8Li
thua
nia
33
−N
ewZ
eala
nd41
993
Mal
aysi
a41
470
923
Nor
way
4277
11M
exic
o22
47−
Por
tug a
l29
62−
Mor
occo
22
1S
inga
pore
188
325
16Pa
kist
an4
41
Sou
thK
orea
7410
023
Per
u5
12−
Spa
in55
105
1P
hilip
pine
s9
15−
Sw
eden
4877
18P
olan
d89
133
2S
witz
erla
nd47
813
Rom
ania
77
1Ta
iwan
290
371
49S
lova
kia
12
−U
.K.
3674
8S
love
nia
11
−U
.S.
210
415
23S
outh
Afr
ica
4868
21S
riLa
nka
11
−T
haila
nd62
100
14T
urke
y62
766
Ven
ezue
la2
4−
Tota
l2,
003
3,50
446
6To
tal
1,30
02,
079
105
Ave
rage
7713
520
Ave
rage
4369
10
(co
ntin
ue
d)
3952
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
Tabl
e2
Con
tinue
d
Pan
elB
:Ave
rage
Dai
lyN
ews
Cov e
rage
Dev
elop
ed[−
55, −
2][ −
1,2]
Em
ergi
ng[ −
55, −
2][ −
1,2]
SM
LS
ML
SM
LS
ML
Aus
tral
ia0.
320.
452.
752.
284.
6413
.03
Arg
entin
a0.
180.
450.
780.
822.
563.
54A
ustr
ia0.
672.
142.
432.
388.
389.
33B
razi
l0.
000.
430.
370.
501.
791.
24B
elgi
um0.
440.
512.
532.
963.
407.
29C
hile
0.50
0.39
1.22
1.54
1.94
2.97
Can
ada
0.42
0.28
1.96
2.00
1.85
8.13
Chi
na0.
080.
280.
720.
581.
683.
59D
enm
ark
0.36
0.48
1.77
3.07
3.73
10.8
2H
ung a
ry0.
330.
600.
941.
442.
752.
71F
inla
nd0.
220.
341.
421.
151.
484.
20In
dia
0.34
0.58
1.45
1.25
2.06
3.86
Fra
nce
0.21
0.54
1.04
0.86
1.66
4.45
Indo
nesi
a0.
060.
220.
690.
500.
812.
13G
erm
any
0.65
0.79
2.51
1.07
4.22
9.86
Mal
aysi
a0.
060.
170.
490.
520.
972.
09G
reec
e0.
030.
030.
530.
340.
321.
40M
exic
o0.
131.
391.
110.
253.
794.
16H
ong
Kon
g0.
070.
140.
530.
561.
314.
24P
olan
d0.
330.
541.
540.
992.
003.
69Ir
elan
d0.
010.
680.
990.
551.
804.
71S
outh
Afr
ica
0.15
0.11
0.58
0.79
0.88
2.43
Isra
el0.
110.
250.
300.
710.
961.
75T
haila
nd0.
180.
351.
110.
701.
362.
79Ita
ly1.
051.
621.
373.
344.
054.
15T
urke
y0.
080.
621.
050.
471.
053.
11Ja
pan
0.12
0.13
0.32
0.82
1.04
1.53
Oth
erE
mer
g.0.
260.
630.
821.
091.
552.
56N
ethe
rland
s0.
280.
751.
542.
144.
677.
85N
ewZ
eala
nd0.
370.
551.
592.
193.
355.
57N
orw
ay0.
410.
721.
393.
002.
904.
97P
ortu
gal
0.42
0.75
1.13
3.61
2.59
3.39
Sin
gapo
re0.
110.
260.
940.
891.
604.
01S
outh
Kor
ea0.
040.
070.
450.
940.
381.
38S
pain
0.39
0.65
2.24
1.56
2.13
6.38
Sw
eden
0.34
0.79
1.57
2.33
4.20
7.10
Sw
itzer
land
0.11
0.76
1.37
2.15
5.93
7.29
Taiw
an0.
140.
180.
680.
810.
962.
26U
.K.
0.85
1.30
0.88
2.95
4.53
5.52
U.S
.0.
220.
330.
701.
101.
793.
45
(co
ntin
ue
d)
3953
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
Tabl
e2
Con
tinue
d
Pan
elB
:Ave
rage
Dai
lyN
ews
Cove
rage
Dev
elop
ed[ −
55, −
2][ −
1,2]
Em
ergi
ng[ −
55, −
2][ −
1,2]
SM
LS
ML
SM
LS
ML
Avg
.0.
320.
601.
341.
762.
845.
540.
190.
480.
920.
821.
802.
92Lg
-S
m1.
023.
780.
732.
10( t
-sta
t.)(8
.17)
(7.0
5)(1
0.25
)(9
.20)
Dev
.-E
mg.
0.13
0.11
0.42
0.94
1.04
2.62
(t-s
tat.)
(2.0
7)(0
.89)
(2.4
9)(4
.24)
(2.4
1)(4
.12)
Dev
.E
mg.
Dev
.–E
mg.
SM
LS
ML
SM
L
[-1,
2]-[
-55,
-2]
1.44
2.24
4.20
0.63
1.32
2.00
0.81
0.93
2.20
(t-s
tat.)
(8.7
7)(7
.51)
(8.7
6)(8
.68)
(8.1
7)(1
2.44
)(4
.53)
(2.7
3)(4
.35)
Pane
lAre
port
sfir
man
dev
entc
ount
s.T
heea
rnin
gsan
noun
cem
ents
ampl
efo
rK
orea
cont
ains
even
tsfr
omF
ebru
ary
2001
toJa
nuar
y20
08.T
heea
rnin
gsan
noun
cem
ents
ampl
efo
ral
loth
erco
untr
ies
cont
ains
even
tsfr
omJa
nuar
y20
04to
Mar
ch20
08.T
hem
erge
ran
noun
cem
ents
ampl
eco
ntai
nsev
ents
from
Janu
ary
2001
thro
ugh
Apr
il20
07.E
arni
ngs
anno
unce
men
tsar
eda
tes
from
Blo
ombe
rgth
atpa
ssed
anau
tom
ated
date
-che
ckin
gpr
oces
s.T
hepr
oces
sco
nsid
ers
anea
rnin
gsan
noun
cem
ent
date
from
Blo
ombe
rgac
cura
teif
itis
with
in+
/−
3da
ysof
aF
activ
aar
ticle
that
mee
tsth
efo
llow
ing
crite
ria:
1)th
efir
m’s
nam
eis
inth
ehe
adlin
eor
lead
para
grap
h;2)
the
artic
leis
tagg
edby
Fac
tiva
asea
rnin
gsne
ws
(tag
c151
);3)
the
head
line
cont
ains
ach
arac
ter
strin
gin
dica
tive
ofan
earn
ings
anno
unce
men
tart
icle
;and
4)th
ere
isno
artic
lem
eetin
gcr
iteria
1–3
inth
epr
ior
60da
ys.S
eeO
nlin
eA
ppen
dix
Afo
rm
ore
deta
ilson
the
auto
mat
edch
ecki
ngpr
oced
ure.
The
sam
ple
isth
eun
ion
ofth
esu
bset
ofF
activ
aco
nfirm
edea
rnin
gsan
noun
cem
ents
with
retu
rns
that
pass
our
filte
rsan
da
set
ofha
nd-c
heck
edK
orea
nda
tes.
The
sam
ple
ofm
erge
ran
noun
cem
ents
isco
mpo
sed
ofth
eea
rlies
tda
tefr
omon
eof
thre
eso
urce
s:S
DC
,F
actS
et,
and
Blo
ombe
rg.
We
rest
rict
the
sam
ple
toin
itial
bids
(no
bids
for
the
targ
etin
the
prio
rtw
oye
ars)
and
mer
gers
whe
reth
eta
rget
has
atle
ast
one
artic
lew
ithth
efir
m’s
nam
ein
the
head
line
orle
adpa
ragr
aph
betw
een
60ca
lend
arda
ysan
dtw
otr
adin
gda
yspr
ior
toth
ean
noun
cem
ent
and
noar
ticle
with
am
erge
rke
ywor
din
the
sam
etim
efr
ame.
The
lang
uage
-spe
cific
mer
ger
keyw
ords
are
Goo
gle
tran
slat
ions
ofac
quire
,ac
quire
s,ac
quis
ition
,bi
d,bi
ds,
buyo
ut,
deal
,m
erge
,m
erge
r,se
ll,ta
keov
er,
and
talk
s.S
peci
alch
arac
ter
sets
prev
ente
dus
from
tran
slat
ing
Chi
nese
,Ja
pane
se,
and
Rus
sian
mer
ger
keyw
ords
,so
we
excl
ude
mer
gers
from
Chi
na,
Japa
n,an
dR
ussi
a.F
orm
ost
lang
uage
s,w
est
emw
ords
usin
gth
eS
now
ball
algo
rithm
s(
http
://sn
owba
ll.ta
rtar
us.o
rg)an
dco
mpa
rero
otw
ords
rath
erth
anth
een
tire
char
acte
rst
ring.
The
rew
ere
nost
emm
ers
avai
labl
efo
rIta
lian,
Hun
garia
n,P
olis
h,T
urki
sh,
orS
lova
k,so
for
thes
ew
ese
arch
edar
ticle
sfo
rth
efu
lltr
ansl
ated
mer
ger
keyw
ords
.P
anel
Bre
port
sth
eav
erag
enu
mbe
rof
artic
les
per
day
for
afir
mdu
ring
[−
55,−
2]an
d[−
1,2]
trad
ing-
day
win
dow
sar
ound
earn
ings
anno
unce
men
ts.T
heS
,M
,an
dL
grou
ping
sst
and
for
smal
l-,m
ediu
m-,
and
larg
e-si
zepo
rtfo
lios
form
edus
ing
prio
r-D
ecem
ber
NY
SE
/NA
SD
AQ
/AM
EX
brea
kpoi
nts.
InP
anel
B,c
ount
ries
with
few
erth
an20
firm
-yea
rob
serv
atio
nsar
egr
oupe
din
toth
eca
tego
ry“O
ther
Em
erg.
”be
fore
aver
agin
g.
3954
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
event across countries. Here, we first outline our methodology for examiningearnings-event reactions.
3.1 PreliminariesAlthough unexpected positive earnings news is typically accompanied by apositive stock price reaction, the market may form expectations differentlyacross markets. Our concern is whether or not information release is con-centrated around news events, and not whether the information is positive ornegative. Hence, we focus our analysis on volatility around the announcement.We take the absolute value of a stock’s return in excess of its local value-weighted market return as our measure of abnormal volatility.14
Normalizedvolatility is the average abnormal volatility during the eventwindow divided by the average abnormal volatility during the 55 days beforeand 55 days after the event minus one. It has intuitive appeal in that it measuresabsolute event returns in proportion to absolute returns outside the earningswindow. If the two periods are equivalent, the ratio will take a value ofzero. Differenced volatility is the average abnormal event volatility minus theaverage abnormal volatility in the±55 days around the event. To ensure thatour country-level findings are not driven by imprecise measurement, for mostof our analysis we include countries with at least 20 events, leaving us with 13emerging and 26 developed markets with more than 20 events. We group the17 emerging markets with fewer than 20 earnings announcement events intoan “Other Emerging” category.
3.2 Event reaction resultsPanel A of Figure1 ranks all of the countries from highest to lowest basedon the event volatility ratio. Statistical significance (denoted by stripes) isdetermined with a non-parametrict-test at the five-percent level.
First, the figure highlights how event reactions vary widely around theworld. Denmark, the U.K., Sweden, the Netherlands, the United States,Finland, Hong Kong, and Germany have event reactions over 0.5, meaningthat volatility during the four-day event window is 50% greater than normalstock volatility. Thailand, Turkey, Mexico, Indonesia, Poland, Argentina,and China exhibit event reactions less than 0.15, indicating that volatilityon earnings announcement days is similar to volatility on days with noearnings news. Second, developed markets in black are typically on theleft side of the graph with much higher event reactions, while emergingmarkets in gray are to the right with low event reactions. The top 18 eventreactions are developed markets. These markets are followed by some ofthe more established emerging markets, India, South Africa, Brazil, andChile, and another group of developed markets, Taiwan, Portugal, Austria,
14 For individual stocksGriffin (2002) shows that bench marking with a local factor model yields more accurateexpected return estimates than a global or international model.
3955
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
The Review of Financial Studies / v 24 n 12 2011
Figure 1(Continued)
3956
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
How Important Is the Financial Media in Global Markets?
Figure 1Earnings announcement reactionsThis figure plots three types of earnings announcement reaction volatility: Normalized Volatility, DifferencedVolatility, and Residual Differenced Volatility. Normalized Volatility in Panel A and Differenced Volatility inPanel B are calculated as in Table3. Panel C aggregates the firmi and yeart residual reactions from regressionsweighted by the reciprocal of the number of events in the country, so as not to over-weight countries with morefirms:
V olEventi,t − V olNormal
i,t = α + I ndustryi + β1ln(Prior Dec.U SD MVi,t
)
+β2Prior Y ear Pct. Price Tradei,t + β3Accuracycountry
+Prior News Coveragei,t .
V olEventi,t is the mean absolute abnormal return over the [-1, 2] event window relative to the earnings
announcement date.V olNormali,t is the mean absolute abnormal return during the [-55, -2] and [3, 55] windows
calculated as described in Section3.1.Industryis a dummy variable for the firm’s Fama and French 17 industry.Prior Dec. USD MVis the prior December-end market value in U.S. dollars.Prior Year Pct. Price Tradeisthe percentage of trading days with non-zero price changes during the prior calendar year.Accuracyfor eachcountry is reported in Online Appendix Table IA.5.Prior News Coverageis the number of Factiva articles in the[-55, -2] window. In Panels A and B, countries with event volatility significantly greater than normal volatilityas indicated by aCorrado(1989) non-parametric rankt-test are denoted by striped bars. Countries with fewerthan 20 firm-year observations are grouped into the category “Other Emerg.”
Italy, South Korea, Greece, and Canada. Interestingly, Canada seems outof place relative to its advanced economy and stock market; however,Bhattacharya(2006) argues that security law enforcement is extremely laxin Canada relative to the United States. Third, the bottom 11 countrieswith low event reactions are emerging markets, except for Spain. The
3957
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
non-parametrict-statisticsfor the figure are also reported in the last columns ofTable3 and show that there is no significant difference between event-day andnon-event-day volatility in most of these 11 low-reaction countries. Finally, thelow event reaction in Mexico is remarkably consistent with the findings fromBhattacharya et al.(2000), who find no market reaction to 75 news events inMexico during an earlier time period (1994 to 1997).
One concern is that differences in reactions might be due to differences inthe distribution of firm size across countries. Table3 summarizes normalizedvolatility for firms sorted into three size portfolios. Panel A is for developedmarkets and Panel B for emerging markets. First, there are significant dif-ferences between developed and emerging markets for the two largest sizebins, indicating that firm-size differences are not likely driving the differencesbetween developed and emerging markets. Second, surprisingly, there aretypically relatively small differences in reactions across size portfolios. Third,for robustness we use a standard market model adjustment usingDimson(1979) (infrequent trading corrected) betas. There is little difference betweenthe Dimson market model adjusted volatility ratios and those calculated usinga simple return minus the market. Finally, the difference between developedand emerging markets is relatively large. In developed markets, the averageevent reaction is 0.38, meaning that average daily volatility during the four-dayearnings event window is 38% more volatile than a typical day. In emergingmarkets, event-window volatility is substantially lower—earnings days exhibitonly 15% more absolute return movement than normal days. These findingsindicate that event days have two and a half times more abnormal volatility indeveloped markets.
Since ratios can be inflated by denominators near zero, Panel B ofFigure1 examines the difference between (as opposed to the ratio of) earningsannouncement window absolute excess returns and absolute excess returnsduring normal times. While the precise ordering varies somewhat from PanelA, the general patterns are similar. There are only three emerging marketsamong the top 25 markets, and 11 emerging markets are among the bottom15 markets (counting the “Other Emerging” category as one country). Table3 shows that developed markets exhibit 0.56% more volatility on earningsdays, whereas emerging markets only have an additional 0.23% in excessreturn movement. Conceptually, we prefer the volatility ratio to the volatilitydifference, because the ratio makes for more intuitive comparisons of abnormalvolatility across countries with different levels of idiosyncratic volatility.
The differences in average country reactions could be due to differences infirm characteristics across countries, such as industry differences in the uncer-tainty and importance of earnings. We estimate a cross-sectional regression atthe firm level to remove these effects and plot residuals from these regressionsat the country level in Panel C of Figure1. The ordering changes somewhatfrom Panel A, but is generally similar. Most developed markets are again onthe left side of the graph with much higher volatility during the announcement
3958
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
Tabl
e3
Ear
ning
sev
entr
eact
ions
Pan
elA
:Dev
elop
edM
ark ets
Nor
mal
ized
Vola
tility
Diff
eren
ced
Vola
tility
(%)
Ran
k-B
ased
t -st
at.
Siz
eP
ortfo
lio(M
kt.A
dj.)
Cou
ntry
-leve
lC
ount
ry-le
vel
Cou
ntry
SM
LM
kt.A
dj.
Dim
.Adj
.M
kt.A
dj.
Dim
.Adj
.M
kt.A
dj.
Dim
.A
dj.
Aus
tral
ia0.
360.
470.
570.
430.
460.
820.
89(3
.69)
(3.9
3)A
ustr
ia0.
340.
150.
210.
200.
230.
280.
34(2
.40)
(2.3
2)B
elgi
um0.
320.
680.
350.
490.
540.
580.
62(6
.07)
(6.3
3)C
anad
a0.
120.
330.
250.
180.
140.
390.
32(3
.50)
(2.4
9)D
enm
ark
0.47
0.54
0.77
0.60
0.61
0.71
0.70
(5.9
9)(5
.91)
Fin
land
0.62
0.54
0.40
0.52
0.61
0.73
0.77
(5.5
5)(6
.18)
Fra
nce
0.28
0.34
0.71
0.38
0.40
0.51
0.53
(3.4
5)(3
.59)
Ger
man
y0.
540.
340.
610.
510.
510.
910.
93(4
.26)
(4.3
6)G
reec
e0.
060.
380.
260.
180.
170.
300.
27(2
.07)
(1.6
7)H
ong
Kon
g0.
550.
550.
310.
510.
510.
920.
94(7
.57)
(7.7
4)Ir
elan
d0.
040.
430.
500.
390.
330.
520.
43(3
.44)
(2.9
5)Is
rael
0.20
0.36
0.27
0.31
0.28
0.37
0.35
(2.7
6)(2
.17)
Italy
0.21
0.14
0.40
0.20
0.20
0.25
0.25
(3.7
1)(3
.37)
Japa
n0.
390.
460.
430.
430.
430.
540.
54(5
.37)
(5.3
0)N
ethe
rland
s0.
360.
600.
700.
560.
580.
700.
72(6
.35)
(6.3
1)N
ewZ
eala
nd0.
520.
360.
470.
440.
410.
470.
46(4
.75)
(4.7
9)N
orw
ay0.
590.
300.
290.
410.
400.
690.
68(4
.46)
(4.5
8)P
ortu
gal
0.13
0.24
0.28
0.21
0.31
0.20
0.29
(3.0
3)(3
.82)
Sin
gapo
re0.
390.
370.
240.
360.
330.
560.
53(5
.72)
(5.2
1)S
outh
Kor
ea−
0.03
0.12
0.25
0.19
0.15
0.36
0.29
(3.2
6)(2
.79)
Spa
in−
0.02
0.05
0.29
0.17
0.17
0.18
0.18
(2.9
3)(2
.86)
Sw
eden
0.54
0.51
0.70
0.56
0.58
0.89
0.90
(4.4
7)(4
.78)
Sw
itzer
land
0.45
0.35
0.53
0.41
0.44
0.51
0.52
(4.7
0)(4
.81)
Taiw
an0.
220.
250.
160.
220.
200.
380.
35(3
.00)
(2.7
3)U
.K.
0.43
0.60
0.67
0.57
0.50
0.84
0.77
(5.8
0)(5
.51)
U.S
.0.
390.
530.
770.
530.
510.
850.
84(8
.02)
(7.7
6)
Dev
.Avg
.0.
330.
380.
440.
380.
380.
560.
55(8
.56)
(8.3
9) (co
ntin
ue
d)
3959
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
Tabl
e3
Con
tinue
d
Pan
elB
:Em
ergi
ngM
arket
s
Nor
mal
ized
Vola
tility
Diff
eren
ced
Vola
tility
(%)
Ran
k-B
ased
t -st
at.
Siz
eP
ortfo
lio(M
kt.A
dj.)
Cou
ntry
-leve
lC
ount
ry-le
v el
Cou
ntry
SM
LM
kt.A
dj.
Dim
.Adj
.M
kt.A
dj.
Dim
.Adj
.M
kt.A
dj.
Dim
.A
dj.
Arg
entin
a−
0.01
0.29
−0.
060.
100.
110.
140.
15(0
.74)
(0.9
7)B
razi
l2.
130.
570.
170.
280.
350.
440.
52(3
.35)
(3.4
3)C
hile
−0.
400.
410.
280.
250.
270.
240.
27(2
.62)
(3.0
7)C
hina
0.00
0.13
0.13
0.12
0.10
0.20
0.18
(1.8
6)(1
.82)
Hun
gary
0.24
0.14
−0.
080.
170.
210.
280.
31(1
.05)
(1.5
2)In
dia
0.36
0.31
0.18
0.30
0.29
0.55
0.55
(5.8
2)(5
.44)
Indo
nesi
a0.
170.
06−
0.12
0.04
0.02
0.07
0.03
(1.1
3)(0
.59)
Mal
aysi
a0.
160.
210.
230.
170.
180.
270.
29(4
.06)
(3.8
9)M
exic
o−
0.63
0.04
0.08
0.04
0.11
0.05
0.14
(0.3
2)(1
.32)
Oth
erE
mer
g.0.
130.
140.
310.
180.
110.
260.
17(1
.90)
(1.6
2)P
olan
d0.
080.
120.
030.
090.
100.
170.
19(2
.43)
(2.3
8)
(co
ntin
ue
d)
3960
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
Tabl
e3
Con
tinue
d
Pan
elB
:Em
ergi
ngM
ark et
s
Nor
mal
ized
Vola
tility
Diff
eren
ced
Vola
tility
(%)
Ran
k-B
ased
t-st
at.
Siz
eP
ortfo
lio(M
kt.A
dj.)
Cou
ntry
-leve
lC
ount
ry-le
v el
Cou
ntry
SM
LM
kt.A
dj.
Dim
.Adj
.M
kt.A
dj.
Dim
.Adj
.M
kt.A
dj.
Dim
.A
dj.
Sou
thA
fric
a0.
220.
340.
280.
290.
300.
400.
42(3
.47)
(3.7
5)T
haila
nd0.
06−
0.06
0.10
0.02
0.02
0.03
0.04
(0.7
7)(1
.22)
Tur
key
0.10
0.03
−0.
210.
020.
080.
040.
13(0
.15)
(1.4
4)
Em
g.A
vg.
0.19
0.19
0.09
0.15
0.16
0.23
0.24
(6.2
0)(6
.32)
Dev
.–E
mg.
0.14
0.19
0.34
0.23
0.22
0.33
0.31
( t-s
tat.)
(0.8
2)(3
.40)
(5.9
7)(5
.93)
(5.4
1)(5
.28)
(4.8
8)
Thi
sta
ble
repo
rts
sum
mar
yst
atis
tics
fort
here
actio
nsto
the
earn
ings
even
tsde
scrib
edin
Tabl
e2.
Rea
ctio
nsar
eba
sed
eith
eron
mar
ket-
adju
sted
abno
rmal
retu
rns
(Mkt
.Adj
.)or
Dim
son-
beta
adju
sted
abno
rmal
retu
rns
(Dim
.A
dj.)
.S
tock
retu
rns
are
mea
sure
din
loca
lcur
renc
y.E
xtre
me
one-
day
retu
rns
over
200%
and
retu
rns
that
reve
rtqu
ickl
yw
itha
one-
day
abso
lute
retu
rnov
er10
0%an
da
two-
day
retu
rnof
less
than
20%
are
elim
inat
edto
redu
cepo
tent
iald
ata
erro
rs.
Inor
der
for
anev
ent
tobe
incl
uded
,th
ere
mus
tbe
atle
ast
50no
n-m
issi
ngre
turn
sin
the
perio
d25
0to
126
days
befo
reth
eev
entf
orth
epu
rpos
esof
mar
ketm
odel
estim
atio
n.B
etas
are
calc
ulat
edfo
llow
ing
Dim
son
(197
9),u
sing
thre
ele
ads
and
lags
ofth
eva
lue-
wei
ghte
dlo
cal
mar
ketr
etur
n.T
here
actio
nst
atis
tics
for
each
coun
try
jar
eco
mpo
sed
oftw
opa
rts:
even
tvol
atili
tyan
dno
rmal
vola
tility
.Eve
ntvo
latil
ityis
the
mea
nab
solu
teab
norm
alre
turn
calc
ulat
edas
desc
ribed
inS
ectio
n3.1
over
the
[ −1,
2]ev
entw
indo
wre
lativ
eto
the
earn
ings
anno
unce
men
tdat
e,
Vol
Even
tj
=1 N
N ∑ i=1
1 4
2 ∑ t=−
1
|AR
i,t|
,
whe
rei
isth
eev
ent,t
isth
eev
entd
ay,a
ndNis
the
num
ber
ofev
ents
inea
chm
arke
t.N
orm
alvo
latil
ityis
the
mea
nab
solu
teab
norm
alre
turn
durin
gth
e[
−55
,−2]
and
[3,5
5]w
indo
ws,
Vol
Nor
ma
lj
=1 N
N ∑ i=1
1 107
−
2∑
t=−
55
∣ ∣ AR
i,t∣ ∣ +
55 ∑ t=3
∣ ∣ AR
i,t∣ ∣
,
whe
rei
isth
eev
enta
ndt
isth
eev
entd
ay.N
orm
aliz
edVo
latil
ityis
aver
age
even
tvol
atili
ty,
Vol
Even
tj
, div
ided
byav
erag
eno
rmal
vola
tility
,Vol
Nor
ma
lj
,m
inus
one.
Diff
eren
ced
Vola
tility
isev
entv
olat
ility
min
usno
rmal
vola
tility
.Siz
epo
rtfo
lios
are
base
don
prio
rD
ecem
ber
NY
SE
/AM
EX
/NA
SD
AQ
terc
ilebr
eakp
oint
sfr
om20
00to
2006
.The
last
two
colu
mns
repo
rtno
n-pa
ram
etric
rank
-bas
edt-s
tats
for
each
coun
try
calc
ulat
edfo
llow
ingC
orra
do(1
989)
for
test
sof
the
hypo
thes
isth
atev
ent-
perio
dvo
latil
ityeq
uals
norm
al-p
erio
dvo
latil
ity.
The
aver
age
atth
ebo
ttom
ofea
chpa
neli
sth
eav
erag
eof
even
trea
ctio
nsw
ithin
eith
erde
velo
ped
orem
ergi
ngm
arke
ts.C
ount
ries
with
few
erth
an20
firm
-yea
rob
serv
atio
nsar
egr
oupe
din
toth
eca
tego
ry“O
ther
Em
erg.
”be
fore
aver
agin
g.
3961
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
window. The five highest-reaction countries are Germany, Hong Kong, theUnited States, Sweden, and the U.K., and the five lowest-reaction countriesare Indonesia, Poland, Thailand, Mexico, and Turkey.
We Winsorize firm-level turnover data at the first and 99th percentiles andexamine turnover before and after earnings announcements for the low- andhigh-earnings-reaction countries for robustness. Confirming inferences fromabsolute returns, turnover spikes on earnings announcements in high-earnings-announcement-reaction countries, but the relative increase is much smaller forlow-reaction countries.15
Overall, relative to normal absolute returns, earnings announcement reac-tions show a striking tendency to be two and a half times higher in developedmarkets. The cross-country differences also indicate interesting variation forcross-country analysis.
4. General News Announcements and Stock Returns
A natural question is whether the patterns we observe around earningsannouncements generalize to the typical news release. We examine the extentto which news-media releases affect returns. After aligning news according toa market’s time zone, we start by examining all news that occurs after the priorday’s close and before the current day’s open.
4.1 Firm-level regressionsBecause the influence of news on prices may differ on a firm-by-firm basis, weestimate the following simple firm-level regressions:
ln(1 +
∣∣ARi,t
∣∣) = αi + βNewsDay,i NewsDayi,t−1
+ βArtCount,i Ar tCounti,t−1 + εi,t , (1)
where the dependent variable is the natural log of one plus the absoluteabnormal return,NewsDayt−1 indicateswhether there is at least one articlewith the firm’s name in the headline or lead paragraph appearing just priorto that trading day, andArtCountt−1 is the number of articles meeting thiscriteria. News days with more articles may be events of greater importance, orthe market may respond to a greater extent simply because more investors areaware of the news.16
TheadjustedR2 from these regressions give a simple sense for how closelyreturns are related to variation in news coverage. We estimate this regressionat the firm level and then average the adjustedR2 acrossall firms in a country.Figure 2 reports the average news adjustedR2 for each country with 26
15 Theseresults are presented in Online Appendix Figure IA.2. For the rest of the article, we focus on absolutereturns because we believe Datastream’s volume data are less accurate than return data.
16 Roll (1988)does not run such regressions with absolute returns but does show that return volatility is slightlylower when public news days are excluded from a sample of U.S. stocks.
3962
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
How Important Is the Financial Media in Global Markets?
Figure 2(Continued)
3963
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
The Review of Financial Studies / v 24 n 12 2011
Figure 2Adjusted R2 from regressions of returns on newsThis figure plots average adjustedR2 from firm-level regressions of returns on news variables. Details of thecalculation of the average adjustedR2 are discussed in Table4 and Section4.1. In Panel A and the left panel ofPanel C, the charts include only “Pre-Trade News” articles with time stamps that occurred between the previousday’s market close and the current day’s market open. In Panel B and the right panel of Panel C, we count allarticles as news even if there is no time stamp as in Panel C of Table4. Emerging markets with fewer than 20firm years are grouped together in the “Other Emerg.” category. Panel C presents the developed- and emerging-market average of the country-level average Adj.R2 measures. Averages are also presented for firms in three“news bins”: 5 to 15 news days in a year, 15 to 25, and 25 and more. Before averaging across countries, wedelete a news-bin observation if there are fewer than five firm-year observations in the bin. In Panels A and B,stripes indicate that the Adj.R2 is significantly higher, using a two-sided bootstrappedt-test withalpha=0.05.
developed markets in black and 25 emerging markets in gray.17 Bars are stripedif the average adjustedR2 is significantly higher than a random reassignmentof (the same number of) news days, using a two-sided bootstrappedt-testwith alpha = 0.05. Countries are ordered according to the magnitude of theadjustedR2. Panel A in Figure2 shows that most developed markets are tothe left with a greater fraction of return volatility “explained” by news daysand article counts on those news days. The average adjustedR2 is over 3%in Denmark, Sweden, the United States, the Netherlands, the U.K., Portugal,and Belgium, whereas Peru, Argentina, Kenya, Pakistan, Latvia, Morocco, thePhilippines, Taiwan, Egypt, and Slovenia exhibit adjustedR2s below 0.005.All emerging markets have average adjustedR2s below 0.014. The developedmarkets of Cyprus, Australia, Canada, Greece, and Taiwan do not conform tothe developed/emerging split as they have adjustedR2s below 0.01.
Corresponding to Figure2, Panel A1 of Table4 presents more detail anddeveloped and emerging-market averages. Panel A1 shows that in developedmarkets, the average adjustedR2 is 0.024, whereas in emerging marketsit is only 0.007, a significant difference using a bootstrapped test; thet-statistic is 5.85. We also ask whether the regression fit is improved on anindividual firm basis. The news variables help explain volatility for 40% ofdeveloped market firms, but only 18% of emerging market firms.
17 We require a minimum of 20 firm-year observations for the country to be included. Six emerging markets withless than 20 firm years are in the “Other emerging” group and count as one of the 25 markets.
3964
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
Tabl
e4
New
sre
gres
sion
fit(a
dj. R
2)
P ane
lA1:
Pre
-Tra
ding
Day
New
sOnl
y
Cou
ntry
Firm
-Yea
rC
ount
Adj
.R2
-A
llN
ews
Impr
oves
Fit
(%of
Firm
s)A
dj.R
2[5
,15)
Adj
.R2
[15,
25)
Adj
.R2
[25,
Inf)
Aus
tral
ia44
0.00
8∗22
.7%
0.00
5∗0.
017∗
0.01
8∗
Aus
tria
660.
021∗
42.4
0.00
20.
032∗
0.02
3∗
Bel
gium
700.
033∗
60.0
0.01
8∗0.
026∗
0.04
2∗
Can
ada
200.
008∗
25.0
––
0.01
1∗
Cyp
rus
210.
009∗
33.3
––
–D
enm
ark
610.
057∗
62.3
–0.
020∗
0.06
5∗
Fin
land
690.
029∗
43.5
0.03
9∗0.
050∗
0.03
4∗
Fra
nce
450.
016∗
37.8
0.01
6∗0.
030∗
Ger
man
y35
0.01
6∗40
.00.
007
0.00
20.
044∗
Gre
ece
570.
006∗
24.6
0.01
5∗–
0.00
8H
ong
Kon
g47
40.
015∗
27.4
0.02
1∗0.
037∗
0.00
8∗
Irel
and
360.
020∗
33.3
0.01
7∗0.
035∗
0.01
9∗
Isra
el53
0.01
4∗30
.20.
005
0.04
0∗0.
028∗
Italy
800.
021∗
35.0
0.00
3–
0.02
5∗
Japa
n15
90.
029∗
50.3
0.03
2∗0.
030∗
0.01
2∗
Net
herla
nds
710.
039∗
46.5
0.02
9∗0.
040∗
0.05
4∗
New
Zea
land
700.
016∗
32.9
0.01
4∗0.
027∗
0.00
5N
orw
ay71
0.02
8∗45
.10.
001
0.02
9∗0.
033∗
Por
tuga
l42
0.03
4∗45
.20.
010∗
0.00
50.
044∗
Sin
gapo
re24
20.
019∗
36.8
0.01
9∗0.
035∗
0.01
9∗
Spa
in86
0.01
7∗33
.70.
014∗
0.00
10.
023∗
Sw
eden
810.
055∗
69.1
0.06
6∗0.
033∗
0.06
7∗
Sw
itzer
land
520.
021∗
40.4
0.00
8∗0.
014∗
0.03
4∗
Taiw
an86
40.
002∗
8.9
0.00
10.
004∗
0.00
9∗
U.K
.37
0.03
5∗59
.50.
033∗
0.04
2∗0.
031∗
U.S
.87
30.
045∗
53.8
0.01
8∗0.
040∗
0.05
8∗
(co
ntin
ue
d)
3965
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
Tabl
e4
Con
tinue
d
Pan
elA
1:P
re-T
radi
ngD
ayN
ews Onl
y
Cou
ntry
Firm
-Yea
rC
ount
Adj
.R2
-A
llN
ews
Impr
oves
Fit
(%of
Firm
s)A
dj.R
2[5
,15)
Adj
.R2
[15,
25)
Adj
.R2
[25,
Inf)
Arg
entin
a13
60.
005∗
16.9
0.00
6∗0.
001
0.00
4∗
Bra
zil
490.
007∗
16.3
0.00
2−
0.00
30.
019∗
Chi
le77
0.00
7∗19
.50.
009∗
0.01
1∗0.
005∗
Chi
na41
80.
008∗
23.2
0.00
9∗0.
005∗
0.01
6∗
Col
ombi
a43
0.01
2∗25
.60.
005
–0.
028∗
Egy
pt86
0.00
24.
7−
0.00
1–
0.01
2∗
Hun
gary
920.
010∗
25.0
0.01
3∗0.
010∗
0.01
3∗
Indi
a44
10.
011∗
26.3
0.01
6∗0.
017∗
0.01
7∗
Indo
nesi
a61
0.00
5∗11
.5−
0.00
10.
007
0.01
2∗
Ken
ya10
00.
005∗
16.0
0.00
8∗0.
016∗
0.00
0La
tvia
300.
003
20.0
0.00
0–
–Li
thua
nia
620.
006∗
14.5
0.00
8∗0.
005
0.00
1M
alay
sia
432
0.01
0∗21
.30.
018∗
0.02
5∗0.
011∗
Mex
ico
520.
014∗
25.0
0.01
4∗0.
008∗
0.02
5∗
Mor
occo
430.
003
11.6
––
–O
ther
Em
erg.
620.
006∗
14.5
0.00
50.
003
0.01
0∗
Paki
stan
127
0.00
4∗17
.30.
004∗
0.00
50.
033∗
Per
u41
0.00
5∗12
.20.
003
0.00
00.
016∗
Phi
lippi
nes
690.
002
7.2
0.00
30.
005
0.00
0P
olan
d13
30.
009∗
20.3
0.00
7∗0.
003
0.01
6∗
Rom
ania
200.
005∗
20.0
––
–S
love
nia
250.
001
4.0
0.00
3–
–S
outh
Afr
ica
145
0.00
9∗24
.10.
012∗
0.01
0∗0.
007∗
Tha
iland
102
0.00
7∗23
.50.
011∗
0.00
50.
004
Tur
key
167
0.01
1∗26
.30.
011∗
0.01
8∗0.
023∗
(co
ntin
ue
d)
3966
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
Tabl
e4
Con
tinue
d
Cou
ntry
Cou
ntry
Cou
ntA
dj.R
2-
All
New
sIm
prov
esF
it(%
ofF
irms)
Adj
.R2
[5,1
5)A
dj.R
2[1
5,25
)A
dj.R
2[2
5,In
f)
Pane
lA1:
Pre
-Tra
ding
Day
New
sO
nly (con
tinue
d)
Dev
elop
ed26
0.02
440
.00.
017
0.02
70.
030
Em
ergi
ng25
0.00
717
.90.
007
0.00
80.
013
Dif
fere
nce
0.01
722
.10.
010
0.01
90.
017
( t-s
tat.)
(5.8
5)(7
.36)
(3.2
0)(5
.70)
(4.6
4)
Pane
lA2:
Pre
-Tra
ding
Day
New
s:S
ampl
ew
ithF
activ
aco
nfirm
edea
rnin
gsan
noun
cem
ents
Dev
elop
ed18
0.04
550
.70.
045
0.04
30.
047
Em
ergi
ng13
0.01
225
.00.
010
0.01
30.
019
Dif
fere
nce
0.03
225
.70.
035
0.03
00.
028
(t-s
tat.)
(4.6
2)(5
.02)
(3.4
8)(4
.60)
(2.8
3)
Pane
lA3:
Pre
-Tra
ding
Day
New
s:E
xclu
ding−
1to
+2
arou
ndea
rnin
gsann
ounc
emen
ts
Dev
elop
ed18
0.04
147
.80.
041
0.03
80.
046
Em
ergi
ng13
0.01
224
.60.
009
0.01
40.
019
Dif
fere
nce
0.02
823
.30.
032
0.02
40.
027
(t-s
tat.)
(4.6
2)(4
.45)
(3.4
0)(3
.37)
(3.0
3)
P ane
lB:L
ocal
Tim
eS
tam
ped
News
Dev
elop
ed26
0.03
943
.00.
025
0.04
10.
043
Em
ergi
ng25
0.01
019
.00.
010
0.01
00.
017
Dif
fere
nce
0.02
924
.00.
015
0.03
10.
026
(t-s
tat.)
(6.3
9)(7
.09)
(2.9
1)(5
.71)
(4.8
7)
Pane
lC:A
llA
vaila
ble
Art
icle
s(n
ottim
ezon
eadju
sted
)
Dev
elop
ed26
0.03
744
.30.
028
0.02
30.
043
Em
ergi
ng29
0.01
324
.00.
007
0.01
40.
019
Dif
fere
nce
0.02
420
.30.
021
0.00
80.
024
(t-s
tat.)
(4.9
1)(5
.07)
(2.5
3)(1
.44)
(4.2
4)
(co
ntin
ue
d)
3967
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
Tabl
e4
Con
tinue
d
Cou
ntry
Cou
ntry
Cou
ntA
dj.R2
-A
llN
ews
Impr
oves
Fit
(%of
Firm
s)A
dj.R
2[5
,15)
Adj
.R2
[15,
25)
Adj
.R2
[25,
Inf)
Pane
lD:D
owJo
nes,
Reu
ters
,F
ina
nci
alT
ime
s,an
dTh
eW
all
Str
ee
tJo
urn
al,
GM
TD
ate Sta
mp
Dev
elop
ed25
0.04
143
.00.
030
0.04
90.
052
Em
ergi
ng23
0.01
423
.20.
015
0.01
40.
023
Dif
fere
nce
0.02
719
.70.
015
0.03
50.
030
(t-s
tat.)
(5.6
9)(5
.54)
(2.5
4)(4
.75)
(4.3
8)
Thi
sta
ble
show
sav
erag
ead
just
edR2
from
firm
-leve
lreg
ress
ions
ofre
turn
son
new
sva
riabl
esde
rived
from
the
artic
les
sum
mar
ized
inTa
ble
1.T
head
just
edR
2co
mes
from
the
follo
win
gre
gres
sion
run
each
firm
year
:
ln( 1
+∣ ∣ A
Ri,
t∣ ∣)=
αi+
βN
ews
Da
y,iN
ews
Da
y i,t−
1+
βA
rtC
ou
nt,i
Art
Co
un
t i,t−
1+
ε i,t
,
whe
re∣ ∣ A
Ri,
t∣ ∣is
firm
i ’s
abno
rmal
vola
tility
onda
ytca
lcul
ated
asde
scrib
edin
Sec
tion3.
1,N
ews
Da
y i,t−
1is
anin
dica
tor
ifth
ere
isan
artic
lew
ithth
efir
m’s
nam
ein
the
head
line
orle
adpa
ragr
aph
onda
yt−1,
and
Ar t
Co
un
t i,t−
1is
the
num
ber
ofth
ese
artic
les.
Pan
elA
only
uses
pre-
trad
ene
ws,
whi
chm
eans
artic
les
that
appe
arfr
omth
ecl
ose
ofth
em
arke
ton
day
t−1
toth
em
arke
top
enon
day t.
Con
sequ
ently
,P
anel
Are
quire
sth
atar
ticle
sha
vetim
est
amps
we
can
conv
ert
tolo
calt
imes
.P
anel
A1
uses
allc
ount
ry-fi
rm-y
ears
.P
anel
A2
pres
ents
aver
ages
over
coun
try-
firm
-yea
rsw
here
we
have
confi
rmed
earn
ings
date
s.P
anel
A3
pres
ents
aver
ages
over
coun
try-
firm
-yea
rsw
here
we
excl
ude
earn
ings
date
s.In
Pan
els
Bth
roug
hD
,w
eus
eco
ntem
pora
neou
sne
ws, N
ews
Da
y i,t
and
Ar t
Co
un
t i,t.
Pan
elB
also
requ
ires
artic
les
have
time
stam
ps,b
utw
eas
sign
artic
les
toda
yt
ifth
eyar
epu
blis
hed
from
the
clos
eon
day
t-1
toth
ecl
ose
onda
yt .P
anel
Cus
esar
ticle
sev
enif
we
dono
thav
eth
eir
loca
ltim
est
amps
and
uses
the
date
publ
ishe
d,re
cord
edin
Gre
enw
ich
Mea
nT
ime,
asth
eda
teof
the
artic
le.P
anel
Dis
the
sam
eas
C,e
xcep
ttha
twe
rest
rictt
hear
ticle
sto
thos
efr
omD
owJo
nes,
Reu
ters
,F
ina
nci
alT
ime
s,orT
he
Wa
llS
tre
etJ
ou
rna
l.T
heto
pof
Pan
elA
1re
port
sav
erag
eA
dj.
R2
forc
ount
ries,
whi
leth
ebo
ttom
ofP
anel
A1
and
Pan
els
A2
thro
ugh
Dre
port
aver
ages
over
emer
ging
and
deve
lope
dm
arke
tsan
dth
eir
diffe
renc
es.C
ount
sfo
rco
untr
ies
inth
eto
pof
Pan
elA
1lis
tthe
num
ber
offir
mye
ars
incl
uded
,but
inth
ere
mai
nder
ofP
anel
A1
and
Pan
els
A2
thro
ugh
D,t
heco
unts
are
the
num
ber
ofco
untr
ies
inth
ecr
oss-
sect
ion.
Adj
.R
2-A
lllis
tsth
eav
erag
eA
dj.R2
for
allfi
rmye
ars.
“New
sIm
prov
esF
it”is
the
perc
enta
geof
firm
year
sw
here
noti
nclu
ding
the
new
s-da
ydu
mm
yan
dar
ticle
coun
tsre
sults
insi
gnifi
cant
lyw
orse
fitba
sed
ona
good
ness
-of-
fitF
-tes
tat
alp
ha
=0.
05.A
dj.R
2-
[5,1
5)is
the
aver
age
Adj
.R2
over
firm
year
sw
ith5
to15
days
with
new
sar
ticle
sdu
ring
the
year
,Adj
.R
2-
[15,
25)
isth
eav
erag
eA
dj.R2
over
firm
year
sw
ith15
to25
new
sda
ys,a
ndA
dj.R2
-[2
5,In
f)is
the
aver
age
Adj
.R2
over
firm
year
sw
ith25
orm
ore
new
sda
ys.T
obe
incl
uded
,eac
hne
ws
coun
tpor
tfolio
mus
tcon
tain
atle
astfi
vefir
mye
ars.
Afir
mye
aris
incl
uded
ifth
ere
are
atle
ast1
00tr
adin
gda
ysw
ithre
turn
s,an
d20
firm
year
spe
rco
untr
y.A
bnor
mal
retu
rns,
AR
i,t,
are
Win
soriz
edat
the
1stan
d99
thpe
rcen
tiles
.Ast
aron
a
coun
try-
leve
lAdj
.R2
inP
anel
A1
indi
cate
sth
atth
eA
dj.R2
issi
gnifi
cant
lyhi
gher
,usi
nga
two-
side
dbo
otst
rapp
edt -te
stw
itha
lph
a=
0.05
,tha
nsp
ecifi
catio
nsw
here
the
sam
enu
mbe
rof
new
sda
ysan
dar
ticle
coun
tsar
era
ndom
lyas
sign
edth
eda
ysin
the
year
.To
calc
ulat
eth
eco
untr
y-le
velb
oots
trap
ped
t -st
atis
ticfo
rea
chof
500
itera
tions
,we
rand
omiz
eth
ene
ws
days
with
inea
chfir
m-y
ear
sam
ple,
estim
ate
the
regr
essi
onde
scrib
edab
ove,
then
calc
ulat
eth
et-
stat
istic
for
the
diffe
renc
ebe
twee
nth
eac
tual
Adj
.R
2an
dth
ebo
otst
rapp
edA
dj.R2.
Sig
nific
ance
isba
sed
onth
edi
strib
utio
nof
thes
eco
untr
y-le
vel
t -st
atis
tics.
Test
sfo
rth
edi
ffere
nce
betw
een
leve
lsin
emer
ging
and
deve
lope
dm
arke
tsar
et -
stat
istic
sbas
edon
boot
stra
pped
stan
dard
erro
rsfr
om1,
000
draw
sw
ithre
plac
emen
tfro
mth
esa
mpl
eof
coun
try-
leve
lave
rage
Adj
.R
2s.
3968
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
Individual articles may have more information content in markets wherethere are fewer articles. Hence, in the last three columns of Table4, wegroup all firms in a market into three bins according to the number of daysin the year that there is news about the firm. The large difference betweendeveloped and emerging markets holds for each news-days-per-year bin.In Panel A2 of Table4, we first limit our sample to the subset of firmswhere earnings announcements have been cross-checked with Factiva newsarticles. In Panel A3, we use this same subset of firms but exclude theearnings announcement window. We still find similar large differences betweenemerging and developed markets.
Rather than only examine pre-trade news, we extend the analysis in PanelB of Table4 to all news over the day (from market close on dayt-1 to theclose on dayt) for the sample of firms (the same as in Panel A) where wehave a time stamp to confirm the exact local time of the article. Next, weexpand the sample in Panel B of Figure2 and Panel C of Table4 even furtherto include all articles, even if we do not have a local time stamp. Both testsshow similar patterns, with most developed markets exhibiting much higheraverage explanatory power in the news regressions. Panel B of Figure2 showsthat financial news bears the strongest relation to prices in northern Europeancountries and the weakest relation in Morocco, Pakistan, Peru, Zimbabwe, andTaiwan.18 Panel C of Figure2 presents developed and emerging averages bybins based on the number of pre-trade news articles in the left panel and onall news in the right panel. In most bins, we find that the explanatory power ofnews for absolute returns is more than twice as high in developed markets. Weexamine news on only those days when the article is published by Dow Jones,The Wall Street Journal, Financial Times, or Reuters in Panel D of Table4 toinvestigate whether the type of news outlet affects inferences. We find similarlarge differences between developed and emerging markets.
Overall, the findings indicate that coverage by the financial media helpsexplain idiosyncratic variation in stock prices much better in developed than inemerging markets.
4.2 RankingThe regressions in the previous subsection are arguably superior to the simplecomparisons between the volatility on news and non-news days we present inthis subsection, because they allow us to model return volatility as a function ofexistence of any news (the news-day dummy) and the importance of the news(the article counts). However, simple comparisons of news-day to non-news-day volatility allow us to more directly quantify the magnitude of increased
18 For Taiwanese earnings announcements, we find that revenues are often released ahead of the earningsannouncement. Thus, one possibility is that the news in the articles is mostly known beforehand through therevenue information.
3969
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
returnvolatility on news days.19 We scale each rank within a country by thenumber of observations to obtain a percentile ranking to make the rankingscomparable across countries, where 0.50 would be the mean percentile rankif news-day and non-news-day volatility were the same. We then calculate theaverage percentile rank and subtract 0.50 to center the averages around zero.
Panel A of Figure3 shows that the distribution of news-day volatility ishigher than non-news-day volatility in most developed markets. Statisticallysignificant shifts in distribution between news-day and non-news-day volatilityare indicated in stripes and are based on the non-parametric Fligner-Policellotest for differences in central tendency. Most developed markets are to the leftof the graph, with relatively higher news-day volatility, and many emergingmarkets to the right, with little difference between news and non-news days.However, there is variation, with Hungary, India, and Thailand exhibitingnews-day volatilities that rank high, whereas the developed markets of Taiwan,Greece, and Cyprus exhibit no meaningful difference between news andnon-news days. There are many emerging markets that show no signifi-cant differences between news days and non-news days (Indonesia, Brazil,Latvia, Romania, the Philippines, Lithuania, Kenya, Pakistan, Morocco, andSlovenia).
In Panel B of Figure3, we report the ratio of the average news-day volatilityto the average non-news-day volatility minus one. In Panel B of Figure3, wesee large differences between news and non-news days. The ratio above 0.70for Japan indicates that the average day with news prior to the market open ismore than 70% more volatile than the average non-news day.20 Overall, thepatterns across countries are similar to Panel A except that more emergingmarkets (Romania, China, Malaysia, Thailand, and India) gravitate to the left,meaning higher reactions, because the aggregation can put considerable weighton extreme observations in a country with few stocks. In Panel C, the medianratio is displayed. Here, the patterns are similar to those in Panel A, with mostdeveloped markets having higher reactions. However, there are still emergingmarkets (Thailand, Romania, India, and Poland) that exhibit higher reactions,whereas a few developed markets (Taiwan, Greece, and Cyprus) have lowermedian reactions. In general, the mean ratios show a slightly more importantrole for news in emerging markets than the medians. This indicates that thereare some large events with news in emerging markets, but in most emergingmarkets the typical news day is no different from a non-news day.
We also examine, for robustness, the developed and emerging averagenews-day rankings, mean and median news to non-news-day volatility ratiosfor samples and classifications of what defines a news day. We examine
19 We start by calculating the average daily excess return volatility for each firm year on news and non-news daysseparately (using news prior to the market opening). Then we pool the average firm-year news and non-newsvolatilities within each country and rank them from lowest to highest average absolute return movement.
20 For significance, we use the two-side 5% level based ont-tests with bootstrapped standard errors.
3970
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
How Important Is the Financial Media in Global Markets?
earnings articles, pre-trading-day news bins, local time-stamped news, allarticles regardless of time stamp, and the major world business news outlets(Dow Jones,The Wall Street Journal, Financial Times, or Reuters) in the
Figure 3(Continued)
3971
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
The Review of Financial Studies / v 24 n 12 2011
Figure 3Pre-trade news-day to non-news-day return volatilityThis figure plots measures that relate average news-day volatility to non-news-day volatility, where volatility,|ARi,t |, is calculated as described in Section3.1. This figure uses only “Pre-Trade News,” as defined in Table4. Volatility is averaged separately for news and non-news days to the firm-year level. In Panel A, we rankaverage firm-year news and non-news volatility from highest to lowest and report the mean rank of the news-day firm-year averages. We divide each rank by the total number of observations to convert the ranks to apercentile and subtract 50 from the percentile for comparability across markets. In Panel B, we average acrossfirm years to get country average news-day and non-news-day volatility, then report the ratio of average news-day to non-news-day volatility minus one. Panel C is the same as B, except using medians. Stripes indicatestatistical significance atalpha = 0.05. In Panels A and B, standard errors for the difference between news-and no-news-day volatility are bootstrapped with 100,000 draws with replacement. In Panel C, the tests areFligner-Policello difference in central tendency tests written in SAS by Paul von Hippel and available athttp://www.sociology.ohio-state.edu/people/ptv/macros/fligner policello.htm.
Online Appendix Table IA.7. Overall, markets tend to move more on newsdays in developed markets than emerging.
5. Investigating Determinants through Cross-country Analysis
Our findings indicate a clear distinction between the market reaction to newsin developed and emerging markets and wide variation within these categoriesthat should be useful for investigating the possible determinants of the return-news relation. We begin investigating hypotheses H1, H2A, H3A, H3B, andH4 through cross-country regressions.
We estimate cross-country OLS regressions with combinations of variablesclosely tied to our hypotheses in Section1. We also examine inferences using aBayesian model selection procedure. The purpose of the Bayesian procedure isto systematically compare our favored hypotheses to alternatives suggested by
3972
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
theliterature. For our purposes, the main advantage of the Bayesian procedureover alternative methods is that it is a robust way to compare a number ofalternative specifications in a small sample.
5.1 Earnings news5.1.1 Basic cross-country regressions.We start by examining nine vari-ables directly linked to our four major hypotheses from Section1. Thevariables that map to these hypotheses are detailed in the Appendix, TableA1. Since we do not have a time series of observations for all explanatoryvariables, we calculate a measure of the average earnings reaction in eachcountry across our four-year sample period. This is used as the dependentvariable. Aggregation across time can reduce noise and does not seem toimpose much cost, given that there is relatively little change in market featuresover our short 2004-to-2008 time period.21
Table5 starts out with nine regressions, each with an intercept and a singlevariable of interest as explanatory variables. Then, as is common in the cross-country literature, we addGDP per capitato capture an alternative null thatour cross-country patterns are simply related to the economic development ofmarkets. The regressions in Table5, Panel A, provide little evidence to supporthypothesis 1: Intra-industry timing of earnings announcements is unableto explain cross-country differences in event reactions. All other variablesare significant in the direction consistent with our hypotheses (HypothesesH2A, H3A, H3B, and H4).Insider trading, technological development, andaccounting standardshave the highestt-statistics and adjustedR2s. Insidertrading, technological development, andaccounting standardsare still highlysignificant, even when includingGDP per capitain the regressions. In supportof the journalism quality hypothesis,news clusteringaround announcementdays andfree pressare also significant, witht-statistics above 2. In short,where insider trading is more prevalent, the price response to earnings news islower. Better news transmission (news clustering, free press, andtechnologicaldevelopment) also yields a stronger response to earnings news.
In Panel B of Table5 we estimate regressions using combinations of twovariables from among those that were significant in Panel A.Insider trading,accounting standards, and technological developmentare highly significant.Free pressis significant in all but one specification withaccounting standards.News clusteringis only significant in one specification. In Panel C, we estimatecombinations of these remaining four significant variables.Insider tradingissignificant in all specifications, andaccounting standardsand technologicaldevelopmentin three of four. Overall, the specifications show considerablesupport for cross-country differences in earnings event reactions being driven
21 We have also collected variables from other papers or from data sources that only cover part of the 2004-to-2008period. In these cases, we use the average value of the data available.
3973
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
Tabl
e5
OLS
regr
essi
ons
ofln
earn
ings
reac
tions
onco
untr
ych
arac
teris
tics
Pan
elA
:Reg
ress
ions
with
One
Varia
ble
ofInte
rest
No
Coe
f.A
dj.R
2G
DP
Coe
f.G
DP
Adj
.R2
Pre
-Eve
ntP
ublic
Info
rmat
ion
Dis
sem
inat
ion
lnA
nn
ou
nce
me
ntO
rde
r−
0.03
−0.
030.
050.
670.
41(−
0.20
)(0
.38)
(5.3
1)In
side
rTra
ding
Insi
de
rT
rad
ing
−0.
790.
62−
0.67
0.17
0.62
(−7.
91)
(−4.
52)
(1.1
2)N
ews
Tra
nsm
issi
onF
ina
nci
alP
ress
Du
rin
gE
ven
t [-1,2
]0.
370.
120.
100.
620.
41(2
.45)
(0.7
0)(4
.45)
New
sC
lust
erin
g0.
510.
240.
280.
540.
47(3
.57)
(2.1
5)(4
.20)
In-d
ep
thA
rtic
les
Du
rin
gE
ven
t [-1,2
]0.
450.
180.
180.
580.
43(3
.04)
(1.2
7)(4
.18)
Fre
eP
ress
0.61
0.36
0.37
0.47
0.51
(4.7
2)(2
.77)
(3.5
2)Te
ch.D
eve
lop
me
nt
0.76
0.57
0.58
0.29
0.61
(7.2
0)(4
.35)
(2.1
4)A
ccou
ntin
gQua
lity
Acc
ou
ntin
gS
tan
da
rds
0.80
0.63
0.63
0.27
0.67
(8.1
4)(5
.39)
(2.3
3)P
ct.I
ntl.
GA
AP
0.40
0.14
0.12
0.62
0.43
(2.6
6)(0
.87)
(4.4
7)C
ontr
ol GD
PP
er
Ca
pita
0.66
0.42
––
–(5
.36)
––
(co
ntin
ue
d)
3974
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
Tabl
e5
Con
tinue
d
Pan
elB
:Reg
ress
ions
with
Two
Varia
bles
ofInte
rest
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
Insi
de
rTra
din
g−
0.43
−0.
65−
0.50
−0.
77(−
3.04
)(−
6.04
)(−
3.78
)(−
5.98
)N
ews
Clu
ste
rin
g0.
040.
170.
280.
29(0
.31)
(1.5
0)(1
.92)
(2.8
3)F
ree
Pre
ss0.
290.
200.
350.
48(2
.67)
(1.6
0)(3
.31)
(3.3
1)Te
ch.D
eve
lop
me
nt
0.40
0.36
0.61
0.67
(2.9
7)(2
.52)
(5.8
6)(6
.48)
Acc
ou
ntin
gSta
nd
ard
s0.
470.
680.
520.
72(3
.30)
(5.5
7)(3
.63)
(6.5
4)
Adj
.R2
0.70
0.67
0.69
0.61
0.65
0.68
0.64
0.66
0.40
0.64
P ane
lC:R
egre
ssio
nsw
ithT
hree
and
Fou
rVa
riabl
esof
Inte
rest
[1]
[2]
[3]
[4]
[5]
Insi
de
rTra
din
g−
0.42
−0.
35−
0.40
−0.
33(−
3.02
)(−
2.38
)(−
3.10
)(−
2.30
)F
ree
Pre
ss0.
180.
250.
220.
20(1
.61)
(2.5
8)(1
.93)
(1.8
5)Te
ch.D
eve
lop
me
nt
0.25
0.36
0.38
0.27
(1.7
4)(2
.88)
(2.7
4)(1
.97)
Acc
ou
ntin
gSta
nd
ard
s0.
370.
340.
370.
22(2
.41)
(2.1
8)(2
.36)
(1.3
1)
Adj
.R2
0.71
0.71
0.73
0.70
0.73
Thi
sta
ble
cont
ains
regr
essi
ons
ofav
erag
ena
tura
llog
ofno
rmal
ized
earn
ings
reac
tions
for
39co
untr
ies
onco
untr
ych
arac
teris
tics.
We
cons
ider
nine
pote
ntia
lexp
lana
tory
varia
bles
that
refle
cthy
poth
eses
abou
tpu
blic
and
priv
ate
pre-
even
tin
form
atio
ndi
ssem
inat
ion,
new
str
ansm
issi
on,
and
acco
untin
gqu
ality
.Va
riabl
esar
ede
scrib
edin
deta
ilin
App
endi
xTa
ble
A1.
All
varia
bles
are
stan
dard
ized
.Pan
elA
pres
ents
regr
essi
ons
with
one
expl
anat
ory
varia
ble.
Pan
elB
pres
ents
regr
essi
ons
with
two
expl
anat
ory
varia
bles
usin
gal
lcom
bina
tions
ofva
riabl
esth
atar
esi
gnifi
cant
inP
anel
Aw
henGD
PP
er
Ca
pita
isus
edas
aco
ntro
l.P
anel
Cpr
esen
tsre
gres
sion
sw
ithth
ree
and
four
expl
anat
ory
varia
bles
usin
gva
riabl
esth
atw
ere
cons
iste
ntly
sign
ifica
ntin
the
Pan
elB
regr
essi
ons.
Par
enth
eses
indi
cate
t -st
atis
tics.
3975
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
by differences in the amount of insider trading, the extent of technologicaldevelopment, and accounting standards.
5.1.2 Stochastic search variable selection.One challenge in the cross-country literature is that there are many explanations for why stocks in onecountry might be different from those in another. Since the sample of countriesis small, the inclusion or exclusion of one variable in a regression may changeinferences about another. A second problem is that for each hypothesis, thereare multiple empirical proxies to choose from. In order to combat these twoproblems and examine the validity of our findings in a systematic frameworkdesigned to deal with such issues, we turn to a well-known Bayesian variableselection procedure.
The Stochastic Search Variable Selection (SSVS) methodology ofGeorgeand McCulloch(1993) embeds standard multiple regression in a hierarchicalBayesian model and is used to identify an important subset of explanatoryvariables. A primary advantage of SSVS over other variable selection methodsis that it selects variables based on the size of their impact on the dependentvariable—their economic significance. The model assumes that the dependentvariable is generated by a combination of the independent variables, and foreach combination, provides a probability that it generated the observed data.These posterior probabilities take into account the number of models andvariables being tested. One is also able to come up with a posterior probabilitythat a variable has a coefficient that is meaningfully different from zero. Weuse these posterior probabilities to identify variables that explain a country’saverage market reaction to news.22
We use a much broader list of variables in the SSVS tests: 33 in theearnings reactions test and 29 in the general news tests. These variableseither relate to our four hypotheses or to more general explanations related toeconomic and financial development, the regulatory environment, governance,or the characteristics of equity markets that are common in the internationalliterature.
SSVS requires priors for the residual variance of the regression,σ 2, thevariances of important and unimportant coefficients,v1 andv0, respectively,and the probability that the independent variables are important,p. We usetwo different priors for the probability that a given independent variable isimportant. The prior ofp = 0.5 says the probability a variable is included in themodel that best explains the dependent variable is the same as the probabilityit is not. This is a standard default prior forp. Since we look at a large numberof variables, we think we should be more skeptical that any given variable is inthe true model. We incorporate this greater skepticism through a second prioron the probability a variable is important,p = 0.15. This more skeptical prior
22 This methodology is similar to the one used byCremers(2002) in selecting important market wide predictorvariables.
3976
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
shrinksposterior probabilities toward zero. Another way to think about this isthat we think there is some probability that a variable looks important in thedata due to random chance, and to account for this, we use a skeptical priorprobability (p = 0.15) that the variables we consider are important. We discussour priors and other details of the SSVS model in Table6.
Panel A of Table6 contains the probability that each independent variableis important for explaining earnings reactions. With the prior ofp = 0.15,financial market sophistication, accounting standards,insider trading, andtechnological developmentare all relatively more important as they have poste-rior probabilities above 0.1. With the less skeptical prior (p = 0.5), these fourvariables plus an investor protection rank and the average number of in-depthnews articles appearing before the event all have posterior probabilities above0.30. Notably, some traditional cross-country determinants, such as insider-trading enforcement, trading costs, and market-level characteristic variables(including market modelR2),23 arenot likely to help explain differences inearnings event reactions.
Panel B of Table6 contains the four best models from each of the two priorsettings, each of those models’ posterior probabilities, along with coefficientestimates and 95% credible intervals.Financial market sophisticationis in thebest model under both priors. With the prior ofp = 0.15,accounting standardsis the second-best model,insider tradingalone third, andfinancial marketsophisticationandinsider tradingtogether are the fourth best model. With theless skeptical (p = 0.5) prior, this model is the third best. The second-bestmodel with the less skeptical prior containsfinancial market sophistication,technological development, the number ofin-depth articles before the event,investor protection, and news clustering. Because we are not exactly surewhat financial market sophisticationis measuring, we take more economiccontent from the other variables. The variable is most highly correlated withaccounting standards,financial disclosure,technological development, andinsider trading(as shown in Online Appendix Table IA.8). The survey variablemay have an advantage over other variables as it captures several facets of theinformation environment related to three of the four main hypotheses.
One concern may be that the cross-country differences are a function ofthe size and industry composition of firms in the market. To control for thesedifferences in Table6, Panel C, we estimate event-level regressions in whichwe first demean each firm’s earnings reaction by size and industry averagereaction.Financial market sophisticationis the most important, followedby the prevalence ofinsider trading. Accounting standardsis the fourthmost important variable, buttechnological developmentis not among the topten. Hence, except fortechnological development, the firm-level regressionsconfirm inferences from the country-level regressions.
23 Kelly (2007) andBartram, Brown, and Stulz(2011) examineR2 andidiosyncratic risk and do not find supportfor it as a measure of information production.
3977
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
Tabl
e6
Reg
ress
ions
ofln
earn
ings
reac
tions
onco
untr
ych
arac
teris
tics
with
Bay
esia
nm
odel
sele
ctio
n
Pan
elA
:Mar
gina
lPos
terio
rP
roba
bilit
yth
atth
eVa
riabl
eis
Impo
rtan
t
Set
tingf
orp
0.15
0.5
0.15
0.5
Fin
an
cia
lMa
rke
tSo
ph
istic
atio
n0.
561
0.69
7P
ct.I
ntl.
GA
AP
0.01
60.
072
Acc
ou
ntin
gS
tan
da
rds
0.26
80.
319
An
ti-S
elf-
De
alin
gIn
dex
0.01
40.
104
Insi
de
rT
rad
ing
0.22
60.
312
Sh
ort
Sa
les
Fea
sib
le0.
014
0.07
8Te
ch.D
eve
lop
me
nt
0.15
80.
324
Ma
rke
tMo
de
lR2
0.01
30.
066
Fin
an
cia
lDis
clo
sure
0.09
60.
196
Fin
an
cia
lPre
ssD
urin
gE
ven
t0.
013
0.09
4In
vest
or
Pro
tect
ion
Ra
nk
0.07
40.
420
Sh
ort
Sa
les
Leg
al
0.01
30.
075
Co
un
try
Ris
k0.
064
0.15
7F
ina
nci
alP
ress
Be
fore
Eve
nt
0.01
30.
082
Fre
eP
ress
0.05
30.
199
lnA
nn
ou
nce
me
ntO
rde
r0.
012
0.06
1In
-de
pth
Art
icle
sB
efo
reE
ven
t0.
039
0.34
3In
sid
er
Tra
din
gE
nfo
rce
d0.
012
0.05
7In
-de
pth
Art
icle
sD
urin
gE
ven
t0.
032
0.20
4D
iscl
osu
reIn
dex
0.01
20.
074
Ma
rke
tTu
rnov
er/
GD
Px
10
00.
031
0.12
4P
ct.D
ays
No
n-Z
ero
Price
Ch
g.0.
012
0.06
9U
.K.L
aw
0.03
00.
143
Dire
cto
rL
iab
ility
Ind
ex0.
012
0.07
6G
DP
pe
rC
ap
ita0.
029
0.11
4C
ost
toE
nfo
rce
Co
ntr
act
s0.
011
0.06
2N
ews
Clu
ste
rin
g0.
028
0.21
5L
OT
Tra
din
gC
ost
0.01
10.
064
Sh
are
ho
lde
rL
aw
suits
Ind
ex0.
020
0.09
0A
vera
geL
ogF
irm
Siz
e0.
010
0.05
7In
vest
or
Pro
tect
ion
Ind
ex0.
018
0.11
2A
vera
geF
irm
-Lev
elP
/E0.
010
0.05
9O
nly
An
nu
alE
arn
.An
n.
0.01
60.
118
(co
ntin
ue
d)
3978
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
Tabl
e6
Con
tinue
d
Pan
elB
:Bes
tMod
els
Set
tingf
orp=
0.15
Set
ting
for p
=0.
50(1
)(2
)(3
)(4
)(5
)(6
)(7
)(8
)
Fin
an
cia
lMa
rke
tSo
ph
istic
atio
n0.
770.
511.
120.
810.
510.
77[.0
51,1
.04]
[0.1
6,0.
86]
[0.8
0,1.
44]
[0.4
5,1.
16]
[0.1
6,0.
85]
[0.5
1,1.
04]
Acc
ou
ntin
gS
tan
da
rds
0.74
[.048
,1.0
0]In
sid
er
Tra
din
g-0
.70
-0.3
8-0
.38
[-0.
96,-
0.45
][-
0.71
,-0.
05]
[-0.
71,-
0.06
]Te
ch.D
eve
lop
me
nt
0.33
[0.0
9,0.
58]
In-d
ep
thA
rtic
les
Be
fore
Eve
nt
-0.3
3-0
.46
[-0.
57,-
0.09
][-
0.70
,-0.
21]
Inve
sto
rP
rote
ctio
nR
an
k0.
410.
38[0
.17,
0.65
][0
.15,
0.60
]N
ews
Clu
ste
rin
g0.
31[0
.08,
0.55
]
P( γ
|Y)
0.20
20.
087
0.04
50.
035
0.00
50.
005
0.00
40.
004
(co
ntin
ue
d)
3979
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
T abl
e6
Con
tinue
d
Pan
elC
:Firm
-Lev
elM
argi
nalP
oste
rior
Pro
babi
lity
that
Top
10Va
riabl
esar
eIm
port
ant
Set
tingf
orp
0.15
0.5
0.15
0.5
Fin
an
cia
lMa
rke
tSo
ph
istic
atio
n0.
120
0.42
7M
ark
etT
urn
ove
r/G
DP
x1
00
0.01
10.
050
Insi
de
rT
rad
ing
0.02
00.
083
Fre
eP
ress
0.00
90.
045
Inve
sto
rP
rote
ctio
nIn
dex
0.01
70.
093
U.K
.La
w0.
008
0.06
1A
cco
un
ting
Sta
nd
ard
s0.
015
0.06
2F
ina
nci
alP
ress
Art
icle
Du
rin
gE
ven
t0.
008
0.04
7S
ha
reh
old
er
La
wsu
itsIn
dex
0.01
30.
092
Fin
an
cia
lDis
clo
sure
0.00
80.
038
Thi
sta
ble
cont
ains
regr
essi
ons
ofth
ena
tura
llog
ofon
epl
usth
eno
rmal
ized
vola
tility
repo
rted
inTa
ble
3fo
r39
coun
trie
son
33po
tent
iale
xpla
nato
ryva
riabl
esde
scrib
edin
Onl
ine
Tabl
eA
.1.A
llva
riabl
esar
est
anda
rdiz
edby
subt
ract
ing
the
cros
s-co
untr
ym
ean
and
divi
ding
byth
est
anda
rdde
viat
ion.
We
use
the
Bay
esia
nS
toch
astic
Sea
rch
Varia
ble
Sel
ectio
nm
etho
dolo
gyof
Geo
rge
and
McC
ullo
ch(19
93):
Y=
Xβ
+ε,
ε∼
N( 0,
σ2
I),
(1)
σ2
∼vλ/χ
2 ν,
(2)
βi|
γi
∼
N( 0,
v 1)w
he
nγi
=1
N( 0,
v 0)w
he
nγi
=0,
(3)
f(γ
)∼∏
pγi i( 1
−p i) 1
−γi.
(4)
The
prio
rsar
eco
ntro
lled
byth
ehy
perp
aram
eter
sλ,ν,v
0,v 1
,an
dp .
One
may
thin
kof
λas
anes
timat
efo
rre
sidu
alva
rianc
ean
dν
asth
esa
mpl
esi
zefo
rth
ere
sidu
alva
rianc
ees
timat
e.W
ech
oose
the
sam
ple
varia
nce
ofth
ede
pend
ent
varia
ble
asou
rva
lue
for
λ,a
ndw
ese
tνat
3.T
hev 0
andv 1
hype
rpar
amet
ers
are
the
prio
rva
rianc
esof
coef
ficie
nts
onun
impo
rtan
tan
dim
port
anti
ndep
ende
ntva
riabl
es,r
espe
ctiv
ely.
We
belie
veth
atan
econ
omic
ally
sign
ifica
ntva
riabl
eis
one
for
whi
cha
one-
stan
dard
-dev
iatio
nch
ange
lead
sto
atle
asta
one-
tent
hst
anda
rd-
devi
atio
nch
ange
inth
ede
pend
entv
aria
ble,
whi
chis
≈[m
ax(N
orm
aliz
edV
ola
tility
)–m
in(N
orm
aliz
ed
Vo
latil
ity)]
/[N
o.o
fCo
un
trie
s].T
his
thre
shol
dco
rres
pond
sto
any
v 0an
dv1
satis
fyin
g0.
12=
log(
v 1/v
0)/
(v−
10
−v−
11
) ,an
dw
eus
ev1/v
0=
3,23
3 .W
eco
nsid
ertw
ova
lues
forp ,
the
prio
rpr
obab
ility
that
anin
depe
nden
tvar
iabl
eis
impo
rtan
t:0.
5an
d0.
15.W
eus
eth
eG
ibbs
sam
pler
toob
tain
2,00
0,00
0dr
aws
from
the
post
erio
rdis
trib
utio
nf(
γ|Y
)af
tera
1,00
0dr
awbu
rn-in
perio
d.P
anel
Are
port
sth
em
argi
nalp
oste
riorp
roba
bilit
yth
atγ
i=
1fo
rall
inde
pend
ent
varia
bles
.In
Pan
elB
,we
iden
tify
the
four
mod
els
with
the
high
estp
oste
rior
prob
abili
tyfr
omea
chof
the
two
prio
rsp
ecifi
catio
nsan
dre
port
mea
nva
lues
ofth
eim
port
antc
oeffi
cien
tsfr
omth
ose
mod
eldr
aws.
The
num
bers
inbr
acke
tsar
e95
%cr
edib
lein
terv
als.
We
also
run
the
proc
edur
eat
the
firm
leve
l,w
ithth
ede
pend
ent
varia
ble
dem
eane
dby
size
and
indu
stry
port
folio
grou
ps.W
eus
eU
.S.s
ize
terc
ilebr
eakp
oint
san
dF
ama
and
Fre
nch
48in
dust
ries.
Pan
elC
repo
rts
post
erio
rpr
obab
ilitie
sfo
rth
eto
p10
varia
bles
from
the
firm
-leve
lres
ults
.
3980
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
5.2 General news regressions and robustnessBeyond earnings news, we examine news more broadly by constructing asimple composite measure of news importance using the standardized averageof the variables from Panel A of Figure2 and the three panels of Figure3.24
We first estimate OLS cross-country regressions in Table7. Only prevalenceof insider trading and technological development survive the simple GDP per-capita controls.
We again apply SSVS to test the robustness of our findings. Panel A ofTable 8 shows that with the prior ofp = 0.15, technological development,the prevalence ofinsider-trading survey variable, and the in-depth articlevariable are the only three variables with posterior probabilities above 0.10.Panel B of Table8 shows that with the prior of 0.15, the favored specificationis technological developmentalone, followed byinsider tradingalone, andthen in the third specification both variables. With the less skeptical prior(p = 0.50), the specification withtechnological development, insider trading,in-depth articles before the event, andfinancial disclosureis the favored speci-fication, followed bytechnological developmentalone, and thentechnologicaldevelopment, insider trading, andfinancial disclosure.
As with the responses to earnings news for robustness, we use this samemethodology at the event level in Panel C of Table8. In these generalnews regressions,financial disclosure,freedom of the press, insider trading,investor protection, andtechnological developmentare the five variables withthe highest posterior probabilities. It is comforting that both the prevalenceof insider tradingand technological developmentare again important in thefirm-level regressions.
Among our earnings reaction and general news regressions, both at themarket and firm level, the prevalence ofinsider tradingis always one of themost important variables.Technological developmentis important in all SSVStests except the firm-level earnings reactions test. Quite sensibly,accountingstandardsare important for understanding the reaction to earnings, but notgeneral news. Overall, the hypothesis with the most support in the data isinsider trading (Hypothesis H2A), followed by news transmission (HypothesisH3B).
6. Implications for Informed and Uninformed Trading
Because our cross-country regressions point to the importance of insidertrading, we seek additional verification by testing Hypotheses H2B and H2C,both of them related to insider trading.
24 Thefour measures areMean Percentile Rank of News Day Volatility, Ratio of Mean News to No-News Volatility,Ratio of Median News to No-News Volatility, andAdj. News R2.
3981
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
Table 7OLS regressions of ln general news on country characteristics
Panel A: Regressions with One Variable ofInterest
No controls GDPcontrol
Coef. Adj. R2 Coef. GDP Adj. R2
Pre-Event Public Information Disseminationln Announcement Order −0.04 −0.02 0.00 0.33 0.17
(−0.31) (0.03) (3.11)InsiderTrading
Insider Trading −0.46 0.39 −0.46 −0.01 0.37(−4.97) (−3.34) (−0.07)
News TransmissionFinancial Press During Event[-1, 2] 0.24 0.08 0.10 0.29 0.19
(2.04) (0.76) (2.41)News Clustering 0.29 0.12 0.16 0.27 0.21
(2.43) (1.22) (2.26)In-depthArticles During Event[-1, 2] 0.12 0.00 −0.05 0.36 0.18
(1.02) (−0.43) (2.96)Free Press 0.23 0.08 0.07 0.30 0.18
(2.02) (0.58) (2.37)Tech. Development 0.47 0.41 0.43 0.05 0.40
(5.18) (3.60) (0.44)AccountingQuality
Accounting Standards 0.36 0.23 0.25 0.18 0.25(3.50) (1.92) (1.40)
Pct. Intl. GAAP 0.21 0.06 0.06 0.31 0.18(1.78) (0.50) (2.54)
ControlGDPPer Capita 0.33 0.20 – – –
(3.17) – –
Panel B: Regression with Two Variables ofInterest
[1]
InsiderTrading −0.24(−1.88)
Tech. Development 0.29(2.20)
Adj. R2 0.45
Thistable contains regressions of a country-level measure of market reaction to news for 38 countries on countrycharacteristics. The measure is the average of four standardized measures: theMean Percentile Rank of NewsDay Volatility, theRatio of Mean News to No-News Volatility, theRatio of Median News to No-News Volatility,and theNews Regression Fit (adj. R2). Construction of the three volatility measures is discussed in Figure3, and Table4 discusses the adj.R2 measure.We consider nine potential explanatory variables which reflecthypotheses about pre-event information dissemination, public and private, news transmission, and accountingquality. Variables are described in detail in Appendix Table A1. All variables are standardized. Panel A presentsregressions with one explanatory variable. Panel B presents one regression with the two explanatory variablesthat are significant in Panel A whenGDP Per Capitais used as a control. Parentheses indicatet-statistics.
6.1 Leakage prior to takeoversAccording to Hypothesis H2B, if the low stock price reaction to news is drivenby insider trading, we should see much more stock price run-up ahead ofmergers in low-reaction markets than in markets where stock prices respondmore to public news. Hence, we separate markets into two groups based onthe ranking of the target firm’s country of origin in the earnings announcement
3982
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
Tabl
e8
Reg
ress
ions
ofth
ege
nera
lnew
sre
actio
nm
easu
reon
coun
try
char
acte
ristic
sw
ithB
ayes
ian
mod
else
lect
ion
Pan
elA
:Mar
gina
lPos
terio
rP
roba
bilit
yth
atth
eVa
riabl
eis
Impo
rtan
t
Set
tingf
orp
0.15
0.5
0.15
0.5
T ech
.Dev
elo
pm
en
t0.
635
0.73
6S
ho
rtS
ale
sFe
asi
ble
0.02
10.
120
Insi
de
rT
rad
ing
0.36
90.
664
Co
stto
En
forc
eC
on
tra
cts
0.01
90.
094
In-d
ep
thA
rtic
les
Be
fore
Eve
nt
0.10
80.
556
An
ti-S
elf-
De
alin
gIn
dex
0.01
70.
124
Co
un
try
Ris
k0.
072
0.25
5D
irect
or
Lia
bili
tyIn
dex
0.01
70.
132
Fin
an
cia
lDis
clo
sure
0.05
30.
440
Inve
sto
rP
rote
ctio
nR
an
k0.
016
0.12
1U
KL
aw
0.03
60.
231
Sh
ort
Sa
les
Leg
al
0.01
60.
096
Dis
clo
sure
Ind
ex0.
033
0.29
7F
ree
Pre
ss0.
016
0.09
8A
vera
geF
irm
-Lev
elP
/E0.
032
0.16
5O
nly
An
nu
alE
arn
.An
n.
0.01
40.
104
Inve
sto
rP
rote
ctio
nIn
dex
0.02
90.
176
Pct
.Da
ysN
on
-Ze
roP
rice
Ch
g.0.
014
0.08
1F
ina
nci
alM
ark
etS
op
his
tica
tion
0.02
70.
156
Ma
rke
tMo
de
lR2
0.01
40.
076
Acc
ou
ntin
gS
tan
da
rds
0.02
70.
143
Fin
an
cia
lPre
ssB
efo
reE
ven
t0.
014
0.08
3M
ark
etT
urn
ove
r/G
DP
x1
00
0.02
50.
207
Insi
de
rT
rad
ing
En
forc
ed
0.01
40.
076
Sh
are
ho
lde
rL
aw
suits
Ind
ex0.
023
0.14
0L
OT
Tra
din
gC
ost
0.01
30.
076
Pct
.In
tl.G
AA
P0.
023
0.12
7A
vera
geL
ogF
irm
Siz
e0.
013
0.09
2G
DP
pe
rC
ap
ita0.
022
0.13
1
Pane
lB:B
estM
odel
s
Set
tingf
orp =
0.15
Set
ting
forp
=0.
50(1
)(2
)(3
)(4
)(5
)(6
)(7
)(8
)
Tech
.Dev
elo
pm
en
t0.
640.
420.
730.
580.
640.
550.
46[0
.35,
0.94
][0
.04,
0.80
][0
.43,
1.02
][0
.24,
0.94
][0
.35,
0.93
][0
.17,
0.93
][0
.11,
0.81
]In
sid
er
Tra
din
g-0
.63
-0.3
5−
0.71
−0.
64−
0.45
[ −0.
93,−
0.33
][-
0.74
,0.0
3][−
1.13
,−0.
31]
[−1.
08,−
0.20
][−
0.82
,−0.
09]
In-d
ep
thA
rtic
les
Be
fore
Eve
nt
−0.
28−
0.33
−0.
34[−
0.54
,0.0
0][−
0.57
,−0.
08]
[−0.
61,−
0.08
]F
ina
nci
alD
iscl
osu
re−
0.49
−0.
51[ −
0.90
, −0.
08]
[ −0.
93, −
0.08
]
P( γ
|Y)
0.30
40.
155
0.02
70.
025
0.00
80.
006
0.00
50.
005
(co
ntin
ue
d)
3983
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
Tabl
e8
Con
tinue
d
Pan
elC
:Firm
-Lev
elM
argi
nalP
oste
rior
Pro
babi
lity
that
Top
10Va
riabl
esar
eIm
port
ant
Set
tingf
orp
0.15
0.5
0.15
0.5
Fin
an
cia
lDis
clo
sure
0.04
60.
182
Dis
clo
sure
Ind
ex0.
014
0.08
1F
ree
Pre
ss0.
024
0.17
0F
ina
nci
alM
ark
etS
op
his
tica
tion
0.01
30.
064
Insi
de
rT
rad
ing
0.02
40.
099
An
ti-S
elf-
De
alin
gIn
dex
0.01
30.
073
Inve
sto
rP
rote
ctio
nIn
dex
0.02
30.
126
Dire
cto
rL
iab
ility
Ind
ex0.
011
0.06
7Te
ch.D
eve
lop
me
nt
0.02
20.
094
Ma
rke
tMo
de
lR2
0.01
00.
056
Thi
sta
ble
cont
ains
regr
essi
ons
ofa
coun
try-
leve
lmea
sure
ofm
arke
trea
ctio
nto
new
sfo
r38
coun
trie
son
coun
try
char
acte
ristic
s.T
hem
easu
reis
the
aver
age
offo
urst
anda
rdiz
edva
riabl
esas
desc
ribed
inTa
ble7.
We
cons
ider
29ex
plan
ator
yva
riabl
es,a
llof
whi
char
est
anda
rdiz
ed.W
eus
eth
eS
toch
astic
Sea
rch
Varia
ble
Sel
ectio
n(S
SV
S)
mod
eldi
scus
sed
inTa
ble
6.P
anel
Are
port
sth
em
argi
nalp
oste
rior
prob
abili
tyth
atea
chva
riabl
eis
impo
rtan
tfor
expl
aini
ngth
eda
ta.I
nP
anel
B,w
eid
entif
yth
efo
urm
odel
sw
ithth
ehi
ghes
tpos
terio
rpr
obab
ility
from
each
ofth
etw
opr
ior
spec
ifica
tions
and
repo
rtm
ean
valu
esof
the
impo
rtan
tco
effic
ient
sfr
omth
ose
mod
eldr
aws.
The
num
bers
inbr
acke
tsar
e95
%cr
edib
lein
terv
als.
We
also
run
the
proc
edur
eat
the
firm
leve
l,w
ithth
ede
pend
ent
varia
ble
dem
eane
dby
size
and
indu
stry
port
folio
grou
ps.
We
use
U.S
.si
zete
rcile
brea
kpoi
nts
and
Fam
aan
dF
renc
h48
indu
strie
s.P
anel
Cre
port
spo
ster
ior
prob
abili
ties
for
the
top
10va
riabl
esfr
omth
efir
m-le
velr
esul
ts.A
ppen
dix
A1
cont
ains
desc
riptio
nsof
allr
egre
ssor
sco
nsid
ered
inth
eS
SV
Spr
oced
ure.
3984
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
How Important Is the Financial Media in Global Markets?
Figure 4Merger buy-and-hold abnormal returns (BHARs) as a percentage of BHARs for the entire periodThis figure shows buy-and-hold abnormal returns (BHARs) for mergers from 55 days before an announcementthrough two days after an announcement as a percent of the BHAR for the entire [-55, 2] period. The sample ofmerger announcements is collected from Bloomberg, Mergerstat, and SDC. We restrict the sample to initial bids(no bids for the target in the prior two years) and mergers where the target has at least one article written about itbetween 60 calendar days and two trading days prior to the announcement and no merger-related articles in thesame time frame. A merger event in any of the three sources is considered the same event as one from anothersource if the bids are for the same target and within two years of each other. We take the earliest announcementdate for each event from the union of all three sources. Events are divided according to whether the firm’s homecountry has an average earnings announcement reaction above or below the median for all countries with atleast 20 earnings announcements and SUEs as calculated in Figure1, Panel A. There are 293 events in the high-reaction group and 278 events in the low-reaction group. Abnormal returns are market adjusted, which meansthey are the buy-and-hold return for a stock minus the buy-and-hold return for the market. In order for an eventto be included, the stock must have 50% of trading days with price changes in the calendar year prior to theevent. The shaded region in each panel marks the [-1, 2] event window. 95% confidence intervals for the BHARsare marked with smaller, lighter circles or triangles.
reactions in Figure1. As described in the data section, we go to great extremesto ensure that the announcement date is the first public announcement datefrom three sources, and there are no news articles that hint of a pending merger.These restrictive criteria leave us with 571 merger events from 34 countries.
Figure4 plots the takeover run-up for the high- and low-earnings-reactioncountries. Although the average price run-up is similar, each group’s pricemove is scaled by the average total run-up for comparison purposes. In high-earnings-reaction countries, the trading ahead of earnings announcements isresponsible for 14% of the pre-announcement run-up by two days prior tothe first public news announcement or rumor date. In contrast, low-earnings-reaction countries have experienced 39% of their price run-up—nearly threetimes the information leakage. The difference in run-up is statistically signifi-cant, with a one-sidedp-value of 0.034. As an additional check on our findings,we relax the requirement that targets have at least one news article in the days
3985
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
prior to the merger, increasing the sample to 1,301 events. The differencesare similar.25 The correlation between event-day volatility around earningsannouncements and merger announcements is 0.44, with at-statistic of 1.88despite having only 17 countries that we can use to calculate the correlation,indicating similar forces driving both events.26
6.2 Implications for reversalsNext, we seek to test the implications for reversals of Hypothesis H2C. Evi-dence in the United States has shown that there are greater reversals followingextreme returns on non-news days than on news days. We hypothesize thatmarkets with more insider trading will have less reversal following extremereturns on non-news days, because markets with more insider trading will haverelatively more informed trading than liquidity trading on non-news days. Forextreme return events, we choose daily returns that are at least two standarddeviations from the prior 250-day mean absolute return. Additionally, to ensureindependence across events, we only use extreme-return events that have noextreme return in the prior 20 days. We separate positive and negative extremereturn events according to whether the country is a high- or low-averageearnings-announcement-reaction country and examine the extent of the returnreversal in the following 20 days.
In high-reaction markets, reversals are present only after non-news days, asshown in Table9. Following news days, reversals are much smaller. Non-news-day reversals are much smaller in low-reaction markets than in high-reactionmarkets. This is prevalent for both positive and negative stock price moves. Forexample, large daily negative stock price moves in high-reaction markets arefollowed by a reversal of 1.31% in high-reaction markets, whereas the reversalis only 0.55 and statistically significantly lower in low-reaction markets. InPanel B, we scale these reversals relative to the size of the initial shock. Thepatterns are similar.27 We also examine reversals more generally across alldays and find that reversals following non-news days are much greater in high-reaction countries than in low-reaction markets (as shown in Online AppendixTables IA.12 and IA.13).
Weaker reversals in low-reaction markets also suggest that there is propor-tionally less liquidity trading in these markets and that reactions to news eventsare important for providing insight into the differences in the magnitude ofreturn reversals across markets. Both the takeover and reversal findings are
25 Theone-sidedp-value is 0.082 and is displayed in Figure IA.3.
26 To reduce noise, we require a country to have at least 10 mergers to be included in this calculation, reducing thenumber of markets to 17. When we repeat the findings in Figure IA.4, Panel B, starting with the larger sampleof 1,301 events, the correlation increases to 0.59, with at-statisticof 3.49.
27 In supplemental results (Panels B1 and B2 of Figure IA.5 and Table IA.11), we sort into three groups based onthe prior percentage of days traded as a liquidity measure. We find that the patterns of no reversals on non-newsdays for stocks in low-reaction markets are prevalent in the most liquid stock group. This is consistent with thepatterns not being driven by illiquidity.
3986
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
Tabl
e9
Buy
-and
-hol
dex
cess
retu
rns
from
days
+1
to+
20fo
llow
ing
extr
eme
retu
rnev
ents
Pan
elA
:Buy
-and
-Hol
dE
xces
s Ret
urns
Ext
rem
eRet
urn
with
No
New
sN
ews
Neg
.Ret
(-)
Pos
.Ret
(+)
Neg
.Ret
(-)
Pos
.Ret
(+)
Hig
hR
eact
ion
1.31
∗−
0.91
∗−
0.33
0.35
Low
Rea
ctio
n0.
55∗
−0.
22−
0.27
0.82
∗
Dif
fere
nce
0.76
−0.
68−
0.06
−0.
48t −
test
2.0
2−
2.2
2−
0.1
2−
1.0
7p -
valu
e(0
.04)
(0.0
3)(0
.90)
(0.2
8 )
Pane
lB:B
uy-a
nd-H
old
Exc
ess
Ret
urns
asa
Per
cent
age
ofD
ayZ
ero
Ret
urns
Ext
rem
eRet
urn
with
No
New
sN
ews
Neg
.Ret
(-)
Pos
.Ret
(+)
Neg
.Ret
(-)
Pos
.Ret
(+)
Hig
hR
eact
ion
−19
.18∗
−11
.33∗
4.19
4.07
Low
Rea
ctio
n−
8.85
∗−
3.18
4.49
12.2
1∗
Dif
fere
nce
−10
.32
−8.
16−
0.31
−8.
14t −
test
−1
.80
−2
.00
−0
.04
−1
.35
p−va
lue
(0.0
7)(0
.05)
(0.9
7)(0
.18)
Thi
sta
ble
pres
ents
buy-
and-
hold
exce
ssre
turn
s(r
etur
nm
inus
the
valu
e-w
eigh
ted
coun
try
port
folio
)fo
llow
ing
days
with
extr
eme
retu
rns.
We
assi
gnda
ysto
bins
depe
ndin
gon
whe
ther
the
extr
eme
retu
rnis
posi
tive
orne
gativ
e,w
heth
erth
est
ock
isfr
oma
high
-or
low
-rea
ctio
nco
untr
y,an
dw
heth
erth
ere
isor
isno
tne
ws
for
the
stoc
kon
the
extr
eme
retu
rnda
y.A
nex
trem
ere
turn
day
isa
day
with
exce
ssre
turn
sat
leas
ttw
ost
anda
rdde
viat
ions
away
from
the
mea
nco
mpa
red
toex
cess
retu
rns
over
the
prev
ious
250
trad
ing
days
.Hig
h-an
dlo
w-r
eact
ion
coun
try
clas
sific
atio
nis
defin
edas
inF
igur
e4.W
ere
quire
alls
tock
sto
have
atle
ast
one
new
sar
ticle
inth
epr
ior
60tr
adin
gda
ysto
ensu
reth
atth
ese
are
firm
sco
vere
dby
our
new
sso
urce
and
noot
her
extr
eme
retu
rns
inth
e20
days
prio
rto
the
obse
rved
extr
eme
retu
rnev
ent
toav
oid
clus
terin
gof
the
even
ts.
Pan
elA
pres
ents
the
raw
day
+1
to+
20bu
y-an
d-ho
ldex
cess
retu
rns
follo
win
gda
ysw
ithex
trem
ere
turn
sin
perc
ent.
Pan
elB
pres
ents
the
aver
age
port
folio
day
+1
to+
20bu
yan
dho
ldex
cess
retu
rns
asa
perc
ento
fthe
aver
age
port
folio
day
zero
retu
rns.
The
diffe
renc
ein
mea
nste
sts
use
apo
oled
/unp
oole
dt-
test
whe
reth
enu
llof
equa
lave
rage
sbe
twee
nem
ergi
ngan
dde
velo
ped
mar
kets
iste
sted
.A
pool
edt-
test
isus
edw
hen
afo
lded
F-t
est
indi
cate
stha
tsam
ple
varia
nces
are
insi
gnifi
cant
lydi
ffere
ntat
the
five-
perc
ents
igni
fican
cele
vel;
othe
rwis
e,an
unpo
oled
t-te
stis
used
.∗in
dica
tess
igni
fican
ceat a
lph
a=
0.05
.
3987
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
consistentwith insider trading driving stock price reactions to news acrosscountries.
7. Conclusion
Despite the fact that developed market firms have more news articles writtenabout them and more days with news coverage, market reactions to theirpublic news events are considerably stronger than for emerging markets.This is surprising given that with less news coverage, one might think thatemerging market news events should be particularly important. Additionally,even among developed or emerging markets, there are large differences inthe extent to which stock prices respond to news. We hypothesize that thedifferences could be due to the extent of public news dissemination beforethe news announcement, insider trading, the quality of the news transmissionmechanism, and the quality of accounting. The prevalence of insider tradingand, to a slightly lesser extent, the quality of news transmission play thestrongest role in explaining cross-market differences in the information contentof financial news. In a sample of mergers carefully screened for prior mergernews, we find that stock price run-up is greater prior to announcements incountries where the response to news is weaker, suggesting substantial privateinformation leakage. Similarly, in markets where public news is less importantfor stock prices, there are fewer reversals following extreme returns on non-news days, which is consistent with relatively more informed and less liquiditytrading on non-news days.
Our cross-country findings may be useful to policymakers, stock exchanges,and investors as they seek to understand differences in the extent to whichmarkets rely on public and non-public information. Even in today’s relativelysophisticated and global capital markets, there appears to be substantial insidertrading in many markets; news from the media is only a partial surprise.Additionally, since understanding the extent of news incorporation and insidertrading has implications for many other aspects of financial markets, wehope our findings will be useful for future academic research to advance ourunderstanding of cross-country differences in trading volume, liquidity, foreign(outsider) ownership, and market valuations.
3988
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
Appendix A:Variable Description
Appendix Table A1Variable descriptionVariable Description
Pre-Event Public Information Disseminationln Announcement Order All U.S. firms and all firms in sample are sorted in to industry according to
Factset’s Primary SIC. All firms from Datastream are matched on ISIN andSEDOL. All firms from CRSP are matched on CUSIP. Each annual earningsannouncement date is compared to all annual earnings announcement datesfrom firms in the same Primary SIC within the prior 120 calendar days. Allannouncements are ranked from first to last and the Announcement Order isthis rank. We take the log of the order and average to the country level incross-country regressions.
Insider TradingInsider Trading The average of the 1999, 2000, and 2002-2003 GCR responses. The question
asks if “Insider trading in your country’s stock markets is (1=pervasive,7=extremely rare).” We reverse this variable so, extremely rare is 1 andpervasive is 7.
Insider Trading Enforced Dummy variable that equals one if insider trading laws exist and are enforcedas of 2008 [fromBhattacharya and Daouk(2002) and updated by authors in2008].
News TransmissionFinancial Press BeforeEvent
Log of one plus the count of articles about the firm inThe Wall Street Journal,Financial Times, Dow Jones, or Reuters in the 55 to 2 days before the event.
Financial Press DuringEvent
Log of one plus the count of articles about the firm inThe Wall Street Journal,Financial Times, Dow Jones, or Reuters in the -1 to 2 days around the event.
In-depth Articles BeforeEvent
The count of articles about the firm in the window 55 to 2 days before theevent with more than 500 words per firm mentioned.
In-depth ArticlesDuring Event
The count of articles about the firm in the window -1 to 2 days around theevent with more than 500 words per firm mentioned.
News Clustering The log ratio of one plus the number of articles per day in the window -1 to 2days around the event to one plus the number of articles per day in the window55 to 2 days before the event.
Free Press Free Pressis from the 2003, 2004, and 2006 GCR and asks if “the mediacan publish/broadcast stories of their choosing without fear of censorship orretaliation.”
Tech. Development Technological Developmentis the average of three variables: TheTechnological Availability measure from 2007, which asks, ”Towhat extent are the latest technologies available in your country?(1 = not available; 7 = widely available)”; the Technological Readinessmeasure from 2006, which asks, ”What is your country’s position intechnology relative to world leaders’? (1 = Behind; 7 = Ahead)”; andTechnological Sophistication from the 2003 GCR, which asks, ”Yourcountry’s position in technology (1=generally lags behind most countries, 7=is among the world’s leaders)”.
Accounting QualityAccounting Standards Average response to question about the strength of accounting standards.
From the 2003, 2004, 2006, and 2007 Global Competitiveness Report, where7 is high/strongly agree and 1 is low/strongly disagree on these surveys.
Financial Disclosure Financial Disclosureis from the 2005-2006 GCR and reports “The level offinancial disclosure required is extensive and detailed.”
Pct. Intl. GAAP Percent of Firms Following International Accounting Standards (IFRS) orU.S. GAAP is the percent of firms following those standards in the sampleof firms included in the country-level averages (as the dependent variable);U.S. firms are assumed to follow GAAP and Worldscope is the source for therest of the world.
Only Annual Earn. Ann. The fraction of firms in the sample that have only annual earnings announce-ments.
(continued)
3989
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
Appendix Table A1ContinuedVariable Description
Firm Follows GAAPor IFRS
An indicator that is one if the firm follows International Accounting Standards(IFRS) or U.S. GAAP. Worldscope is the source for all foreign firms, and U.S.firms are assumed to follow GAAP.
Economic & Financial DevelopmentGDP per Capita ln(GDP per Capita) in 2000 constant dollars is from World Bank’s Financial
Structure Dataset developed byBeck, Demirguc-Kunt, and Levine(2000) andare annual observations averaged over 2003 to 2007.
Market Turnover/GDP x 100
Market Turnover/GDPin 2000 constant dollars is from World Bank’s Finan-cial Structure Dataset developed byBeck, Demirguc-Kunt, and Levine(2000)and are annual observations averaged over 2003 to 2007.
Financial MarketSophistication
The average measure ofFinancial Market Sophisticationfrom the 2003through 2007 Global Competitiveness Report and asks if “The level ofsophistication of financial markets is higher than international norms.”
Ln Firm’s PriorDec. USD MV
Natural log of the firm’s prior December market capitalization in U.S. dollars.
Regulatory EnvironmentShort Sales Legal Dummy variable set to 1 if short sales are not against the law (Charoenrook
and Daouk 2005).Short Sales Feasible Dummy variable set to 1 if short sales are actually used in the market
(Charoenrook and Daouk 2005).UK Law Dummy variable for whether the legal system in a country is based on
common law.Cost to Enforce Contracts Costs to enforce contracts as a percentage of debt. Average of the annual
measures from 2005 through 2008 from Doingbusiness.org.Governance
Investor Protection Rank Average of the annual measures from 2003 through 2008 from Doingbusi-ness.org.
Investor Protection Index Average of the annual measures from 2005 through 2008 from Doingbusi-ness.org.
Anti-Self-Dealing Index FromDjankov et al.(2008).Shareholder Lawsuits Index Average of the annual measures from 2005 through 2008 from Doingbusi-
ness.org.Director Liability Index Average of the annual measures from 2005 through 2008 from Doingbusi-
ness.org.Disclosure Index Average of the annual measures from 2005 through 2008 from Doingbusi-
ness.org.Trading Costs
LOT Trading Cost Computed followingLesmond et al.(1999). We use the prior year averageover firms included in the sample from 2003 through 2008.
Pct. Days Zero Price Chg. Alternate liquidity measure that is percent of days with zero price changes inthe prior year for the firms in the sample from 2003 through 2008.
Characteristics of Equity MarketsAverage Log Firm Size Natural log of the prior December firm size averaged over all firms in our
sample.Average Firm-Level P/E Average P/E ratio for firms in sample. P/E data are from Compustat for the
U.S. and Datastream and World scope for the rest of the world. Values arefrom the previous calendar year.
Market Model R2 Natural log of (R2/(1-R2)) whereR2 is the SST weighted average R2 ofsimplemarket model regressions including the local and the U.S. market foreach stock in our sample for each year from 2003 through 2008 (through 2007in the U.S.) and averaged over all years.
Country Risk Average over the period 2003-2008 of the country risk index published byEuromoney. Higher values indicate lowerrisk.
ReferencesBae, K., W. Bailey, and C. Mao. 2006. Stock Market Liberalization and the Information Environment.Journalof International Money and Finance25:404–28.
Bailey, W., G. Karolyi, and C. Salva. 2006. The Economic Consequences of Increased Disclosure: Evidencefrom International Cross-listings.Journal of Financial Economics81:175–213.
3990
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
HowImportant Is the Financial Media in Global Markets?
Bartram, S., G. Brown, and R.Stulz. 2011. Why Are U.S. Stocks More Volatile? Working Paper, Ohio StateUniversity.
Beck, T., A. Demirguc-Kunt, and R. Levine. 2000. A New Database on the Structure and Development of theFinancial Sector.World Bank Economic Review14:597–605.
Bekaert, G., C. Harvey, and C. Lundblad. 2007. Liquidity and Expected Returns: Lessons from EmergingMarkets.Review of Financial Studies20:1783–831.
Bhattacharya, U. 2006. Enforcement and Its Impact on Cost of Equity and Liquidity of the Market. InCanadaSteps Up: Report of the Task Force to Modernize Securities Regulation in Canada 6(Task Force to ModernizeSecurities Legislation in Canada, Thomas I.A. Allen, chairman).
Bhattacharya, U., and H. Daouk. 2002. The World Price of Insider Trading.Journal of Finance57:75–108.
Bhattacharya, U., H. Daouk, B. Jorgenson, and C. Kehr. 2000. When an Event Is Not an Event: The CuriousCase of an Emerging Market.Journal of Financial Economics55:69–101.
Bhattacharya, U., N. Galpin, X. Yu, and R. Ray. 2009. The Role of the Media in the Internet IPO Bubble.Journalof Financial and Quantitative Analysis44:657–82.
Bris, A., 2005. Do Insider Trading Laws Work?European Financial Management11:267–312.
Bushee, B., J. Core, W. Guay, and J. Wee. 2010. The Role of the Business Press as an Information Intermediary.Journal of Accounting Research48:1–19.
Chan, W. 2003. Stock Price Reaction to News and No-news: Drift and Reversal after Headlines.Journal ofFinancial Economics70:223–60.
Charoenrook, A., and H. Daouk. 2005. A Study of Market wide Short selling Restrictions. Working Paper,Cornell University.
Chipman, H., E. George, and R. McCulloch. 2001. Practical Implementation of Bayesian Model Selection. InModel Selection, ed. P. Lahiri. Beachwood, OH: Institute of Mathematical Statistics, Monograph Series, volume38:65–134.
Corrado, C. 1989. A Non parametric Test for Abnormal Security-price Performance in Event Studies.Journalof Financial Economics23:385–95.
Cremers, M. 2002. Stock Return Predictability: A Bayesian Model Selection Perspective.Review of FinancialStudies15:1223–49.
DeFond, M., M. Hung, and R. Trezevant. 2007. Investor Protection and the Information Content of AnnualEarnings Announcements: International Evidence.Journal of Accounting and Economics43:37–67.
Dimson, E. 1979. Risk Measurement When Shares Are Subject to Infrequent Trading.Journal of FinancialEconomics7:197–226.
Djankov, S., R. La Porta, F. Lopez-de-Silanes, and A. Shleifer. 2008. The Law and Economics of Self-dealing.Journal of Financial Economics88:430–65.
Dyck, A., N. Volchkova, and L. Zingales. 2008. The Corporate Governance Role of the Media: Evidence fromRussia.Journal of Finance63:1093–135.
Fang, L., and J. Peress. 2009. Media Coverage and the Cross-section of Stock Returns.Journal of Finance64:2023–52.
Fernandes, N., and M. Ferreira. 2009. Insider Trading Laws and Stock Price Informativeness.Review ofFinancial Studies22:1845–77.
Foster, G. 1981. Intra-industry Information Transfers Associated with Earnings Releases.Journal of Accountingand Economics3:201–32.
Freeman, R., and S. Tse. 1992. An Earnings Prediction Approach to Examining Intercompany InformationTransfers.Journal of Accounting and Economics15:509–23.
3991
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from
TheReview of Financial Studies / v 24 n 12 2011
Gagnon, L., and G. Karolyi. 2009. Information, Trading Volume, and International Stock Return Comovements:Evidence from Cross-listed Stocks.Journal of Financial and Quantitative Analysis44:953–86.
George, E., and R. McCulloch. 1993. Variable Selection via Gibbs Sampling.Journal of the American StatisticalAssociation88:881–89.
Givoly, D., and D. Palmon. 1982. Timeliness of Annual Earnings Announcements: Some Empirical Evidence.Accounting Review57:486–508.
Griffin, J. 2002. Are the Fama and French Factors Global or Country-specific?Review of Financial Studies10:783–803.
Griffin, J., P. Kelly, and F. Nardari. 2010. Do Market Efficiency Measures Yield Correct Inferences?A Comparison of Developed and Emerging Markets.Review of Financial Studies23:3225–77.
Griffin, J., T. Shu, and S. Topaloglu. 2011. Examining the Dark Side of Financial Markets: Do Institutions Tradeon Information from Investment Bank Connections? Working Paper, University of Texas.
Hou, K. 2007. Industry Information Diffusion and the Lead-lag Effect in Stock Returns.Review of FinancialStudies20:1113–38.
Huberman, G., and T. Regev. 2001. Contagious Speculation and a Cure for Cancer: A Nonevent That MadeStock Prices Soar.Journal of Finance56:387–96.
Kelly, P. 2007. Information Efficiency and Firm-specific Return Variation. Working Paper, University of SouthFlorida.
Lang, M., K. Lins, and D. Miller. 2003. ADRs, Analysts, and Accuracy: Do ADRs Improve a Firm’s InformationEnvironment and Lower Its Cost of Capital?Journal of Accounting Research41:317–45.
Lesmond, D., J. Ogden, and C. Trzcinka. 1999. A New Estimate of Transaction Costs.Review of FinancialStudies12:1113–41.
Lesmond, D. 2005. Liquidity of Emerging Markets.Journal of Financial Economics77:411–52.
Lia, S., L. Ng, and B. Zhang. 2009. Informed Trading around the World. Working Paper, Singapore ManagementUniversity.
Patton, A., and M. Verardo. 2010. Does Beta Move with News? Firm-specific Information Flows and Learningabout Profitability. Working Paper, Duke University.
Roll, R. 1988.R2. Journal of Finance43:541–66.
Tetlock, P. 2007. Giving Content to Investor Sentiment: The Role of Media in the Stock Market.Journal ofFinance62:1139–68.
Tetlock, P., M. Saar-Tsechansky, and S. Macskassy. 2008. More Than Words: Quantifying Language to MeasureFirms’ Fundamentals.Journal of Finance63:1437–67.
Tetlock, P. 2010. Does Public Financial News Resolve Asymmetric Information?Review of Financial Studies23:3520–57.
———. 2011. All the News That’s Fit to Reprint: Do Investors React to Stale Information?Review of FinancialStudies24:1481–512.
3992
by Patrick Kelly on N
ovember 15, 2011
http://rfs.oxfordjournals.org/D
ownloaded from