Financial markets and public information

Storkenmaier, Andreas

2011

KIT Scientific Publishing

The last decades have seen dramatic changes in trading technology and the way that financial markets operate. As trading technology advances, news providers have kept pace and deliver news to market participants around the world within fractions of a second using electronic systems. Currently, most news is still interpreted by humans but news providers have started to offer newswire products with machine learning systems that specifically cater to algorithmic traders. In practice, newswire messagesmake up a major part of the public information set available to investors. This book studies how newswire messages impact modern electronic equity markets.

Business

Management

English

9783866446946

Andreas Storkenmaier

Financial Markets
and Public Information

Andreas Storkenmaier
Financial Markets and Public Information

Financial Markets
and Public Information
by
Andreas Storkenmaier

Dissertation, Karlsruher Institut für Technologie
Fakultät für Wirtschaftswissenschaften
Tag der mündlichen Prüfung: 26. Mai 2011
Referenten: Prof. Dr. Ryan Riordan, Prof. Dr. Rudi Studer

Impressum
Karlsruher Institut für Technologie (KIT)
KIT Scientific Publishing
Straße am Forum 2
D-76131 Karlsruhe
www.ksp.kit.edu
KIT – Universität des Landes Baden-Württemberg und nationales
Forschungszentrum in der Helmholtz-Gemeinschaft

Diese Veröffentlichung ist im Internet unter folgender Creative Commons-Lizenz
publiziert: http://creativecommons.org/licenses/by-nc-nd/3.0/de/

KIT Scientific Publishing 2011
Print on Demand
ISBN 978-3-86644-694-6

Contents
List of Figures

vii

List of Tables

ix

List of Abbreviations

xi

1

Introduction
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1.2 Structure of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2

High-Frequency Market Dynamics and Public Information
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.3 Institutional Details . . . . . . . . . . . . . . . . . . . . . . . . .
2.4 Data and Sample Selection . . . . . . . . . . . . . . . . . . . . .
2.4.1 Stock Market Data . . . . . . . . . . . . . . . . . . . . .
2.4.2 News Data . . . . . . . . . . . . . . . . . . . . . . . . . .
2.4.3 Sample Selection . . . . . . . . . . . . . . . . . . . . . .
2.5 Research Design . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.5.1 Information . . . . . . . . . . . . . . . . . . . . . . . . .
2.5.2 Trading Intensity, Liquidity, and Volatility . . . . .
2.5.3 Returns and Profitability . . . . . . . . . . . . . . . . .
2.6 Results and Interpretation . . . . . . . . . . . . . . . . . . . . .
2.6.1 Information . . . . . . . . . . . . . . . . . . . . . . . . .
2.6.2 Trading Intensity, Liquidity, and Volatility . . . . .
2.6.3 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . .
2.6.4 Returns and Profitability . . . . . . . . . . . . . . . . .
2.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3

Fragmented Markets and Public Information
3.1 Introduction . . . . . . . . . . . . . . . . . . . .
3.2 Related Work . . . . . . . . . . . . . . . . . . .
3.3 Institutional Details . . . . . . . . . . . . . . .
3.4 Data and Sample Selection . . . . . . . . . . .
3.4.1 Stock Market Data . . . . . . . . . . .
3.4.2 News Data . . . . . . . . . . . . . . . .
3.4.3 Sample Selection . . . . . . . . . . . .

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

1
1
3

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

7
7
8
16
17
17
18
21
22
22
27
30
31
38
43
47
55
57

.
.
.
.
.
.
.

59
59
60
65
67
67
68
69

Contents

vi
3.5

.
.
.
.
.
.
.
.

. 70
. 70
. 73
. 75
. 77
. 77
. 80
. 100

Comovement in International Equity Markets and Public Information
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.2 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3 Data and Sample Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.1 Market Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.2 News Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.3 Cross-Sectional Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.4 Sample Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.4 Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.4.1 Stock Market Comovement . . . . . . . . . . . . . . . . . . . . . . . . .
4.4.2 News Comovement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5 Results and Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5.1 Descriptive Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5.2 Influence of News Comovement on Stock Market Comovement .
4.5.3 Cross-Country Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.5.4 US Specific Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

105
105
107
113
113
114
114
118
121
121
123
124
124
128
131
144
145

3.6

3.7
4

5

Measures . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5.1 Spreads and Trading Activity . . . . . . . .
3.5.2 Information Shares . . . . . . . . . . . . . .
3.5.3 Trade and Quote Correlated Information
Results and Interpretation . . . . . . . . . . . . . . .
3.6.1 Descriptive Statistics . . . . . . . . . . . . .
3.6.2 The Effect of Public Information . . . . .
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

Conclusion
147
5.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
5.2 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

A Sample Firms LSE/Chi-X

151

B Sample Data

155

C RNSE File Format

161

References

163

List of Figures
2.1
2.2
2.3
2.4
2.5
2.6

Novel Intraday News Per Year and Month on the TSX 2005 to 2008 .
Novel Intraday News Per Weekday on the TSX 2005 to 2008 . . . . .
Novel Intraday News Per Time of Day on the TSX 2005 to 2008 . . .
Novel Intraday News Per Year and Month on the TSX 2003 to 2006 .
Novel Intraday News Per Weekday on the TSX 2003 to 2006 . . . . .
Novel Intraday News Per Time of Day on the TSX 2003 to 2006 . . .

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

.
.
.
.
.
.

20
21
22
48
49
50

3.1
3.2
3.3

Consolidated Order Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
News Per Month for LSE Listed Stocks . . . . . . . . . . . . . . . . . . . . . . .
News Per Associated Trading Day of the Week . . . . . . . . . . . . . . . . . .

72
81
82

List of Tables
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
2.10
2.11

Sample News . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Descriptive Statistics Sample Companies . . . . . . . . . . . . . . . . . . . . . . .
Descriptive Statistics Market Measures . . . . . . . . . . . . . . . . . . . . . . . .
Information Estimations Around News . . . . . . . . . . . . . . . . . . . . . . .
Trading Intensity and Liquidity Estimations Around News . . . . . . . . . .
Realized Volatility Estimations Around News . . . . . . . . . . . . . . . . . . .
Likelihood Ratio Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Information Estimations Around News – Control Period . . . . . . . . . . .
Trading Intensity and Liquidity Estimations Around News – Control Period
Likelihood Ratio Tests – Control Period . . . . . . . . . . . . . . . . . . . . . . .
Profitability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.1
3.2
3.3
3.4
3.5
3.6
3.7
3.8
3.9
3.10
3.11
3.12
3.13

Sample News . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Descriptive Statistics Spreads . . . . . . . . . . . . . . . . . . . . . . . .
Descriptive Statistics Spreads by Trade Size . . . . . . . . . . . . . . .
Descriptive Statistics Trading Activity . . . . . . . . . . . . . . . . . .
Descriptive Statistics Price Discovery . . . . . . . . . . . . . . . . . . .
Descriptive Statistics News . . . . . . . . . . . . . . . . . . . . . . . . .
Spreads and Public Information . . . . . . . . . . . . . . . . . . . . . .
Spreads in the Consolidated Order Book and Public Information
Effective Spread by Trade Size and Public Information . . . . . . .
Realized Spread by Trade Size and Public Information . . . . . . . .
Price Impact by Trade Size and Public Information . . . . . . . . . .
Trading Activity and Public Information . . . . . . . . . . . . . . . .
Price Discovery and Public Information . . . . . . . . . . . . . . . . .

4.1
4.2
4.3
4.4
4.5
4.6

Descriptive Statistics Thomson Reuters Business Classification . . . . . . . .
Market Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Descriptive Statistics Monthly Comovement . . . . . . . . . . . . . . . . . . . .
Regression of Country Comovement on World Comovement . . . . . . . .
Influence of News Comovement on Stock Market Comovement . . . . . . .
Influence of News Comovement on Stock Market Comovement – ‘Sentiment’ Only . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Descriptive Statistics Countries . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Influence of News Comovement on Stock Market Comovement – Subsamples 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4.7
4.8

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.

32
33
34
36
41
46
47
51
53
56
58

. 83
. 84
. 85
. 86
. 87
. 88
. 89
. 93
. 94
. 95
. 96
. 101
. 103
117
120
126
127
132
133
134
139

List of Tables

x
4.9

Influence of News Comovement on Stock Market Comovement – Subsamples 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.10 Cross-Sectional Correlations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.11 Influence of News Comovement on Stock Market Comovement – USA Only 143
A.1 Chapter 3 Sample Firms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
B.1
B.2
B.3
B.4

Raw TAQ Data – TSX . . . . . .
Raw Depth Level 3 Data – TSX
Raw TAQ Data – Chi-X . . . . .
Raw Daily Data . . . . . . . . . . .

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

.
.
.
.

156
157
158
159

List of Abbreviations
AEX
AMEX
CAC
CPI
DAX
ECN
FSA
GDP
GMM
ICT
IME
IPO
IT
KIT
LSE
MiFID
MRR
MTF
NIST
NMS
NYSE
OLS
PNAC
RIC
RNSE
SEC
SIRCA
SVM
TAQ
TRBC
TSX
VAR
VMA
WFE
WSJ

Amsterdam Exchange Index
American Stock Exchange
Cotation Assistée en Continu
Corruption Perceptions Index
Deutscher Aktien Index
Electronic Communication Network
Financial Services Authority
Gross Domestic Product
General Method of Moments
Information and Communication Technology
Information and Market Engineering
Initial Public Offering
Information Technology
Karlsruhe Institute of Technology
London Stock Exchange
Markets in Financial Instruments Directive
Madhavan, Richardson, and Roomans
Multilateral Trading Facility
National Institute of Standards and Technology
National Market System
New York Stock Exchange
Ordinary Least Squares
Primary News Access Code
Reuters Instrument Code
Reuters NewsScope Sentiment Engine
US Securities and Exchange Commission
Securities Industry Research Centre of Asia-Pacific
Support Vector Machine
Trade and Quote
Thomson Reuters Business Classification
Toronto Stock Exchange
Vector Autoregression
Vector Moving Average
World Federation of Exchanges
Wall Street Journal

Chapter 1
Introduction
1.1 Motivation
“Markets function only through the transmission of information – both good
and bad. It used to be that the fast horse, the clipper ship, or Mister Reuter’s
land telegraph brought the news by which fortunes were made and lost. Today
it is the electron.”1
This quote, given almost thirty years ago, posits a long history of interaction between
news and financial markets while also addressing the dramatic increase in the speed of news
dissemination since the early days of large news organizations. Finance literature includes
studies that analyze news events even back in the mid 19th century. Willard et al. (1996) use
prices from financial markets to find important events as perceived by the public during
the American civil war based on reactions of asset prices to war news. The 19th century
financial markets already incorporated new information from news into security prices.
The study by Willard et al. (1996) analyzes financial markets and public information during
the early days of modern financial markets when information was indeed still transmitted
on a “fast horse, the clipper ship, or Mister Reuter’s land telegraph”.
To make informed decisions on the basis of public information, market participants have
always requested expeditious and accurate news. In a world without seamless electronic
communication news providers already put great effort into delivering business news as
fast as possible. One example is, as already pointed out, the fierce competition for war
1

Walter B. Wriston, “The Information Society: From Gutenberg to SWIFT”, speech given at the
SWIFT Conference SIBOS ’82 on 23 September 1982 in Washington, D.C., Permanent URL:
http://hdl.handle.net/10427/36045. Walter B. Wriston was CEO of Citicorp. from 1967 to 1984.

1 Introduction

2

news during the American civil war which could have had a profound effect on financial
markets. To speed up the delivery of American news to England, Paul Julius Reuter, the
founder of Thomson Reuters2 , had a telegraph line built from the south-west corner of
Ireland to Cork which already had a telegraph connection to London. Selected American
news were put into water-proofed canisters on mail steamers arriving from the United
States and then thrown into to the water close to the south-west Irish coast. A Reuters
steam-tender would then pick up those canisters and the news would be cabled to London,
delivering news substantially faster than other news providers (Read, 1999, p. 40).
As communication technology advanced, financial market participants requested the
same improvement from news delivery. Once the transatlantic telegraph cable was operational, the speed of news between North America and England had dramatically increased.
A long delay of news “had become unacceptable. From now onwards the business community [in London] expected to receive American stock market and commodity information
via Reuters in hours instead of days” (Read, 1999, p. 94). This trend has continued until
today dramatically driven by the recent advancements in communication and information
technology.
The last decades have seen drastic changes in trading technology and the way that financial markets operate. Starting with the computerization of tasks on exchange floors,
over the introduction of completely electronic markets, up to algorithmic trading which
now makes up more than half of equity trading by recent estimates, trading has become
almost completely computerized (Hendershott and Riordan, 2009). Technology and computers have also revolutionized financial news dissemination and created demanding requirements to financial news products from the customer side. As trading technology has
advanced, news providers like Thomson Reuters, Bloomberg, and Dow Jones have kept
pace and deliver news to market participants around the world within fractions of a second through electronic systems. News that could have taken substantial time to reach
financial market centers only a few decades ago is now globally available at the click of a
button. Global data networks and satellites even reach to the remotest places on earth. In
the quote at the beginning of this chapter, Walter B. Wriston talks about “the electron” as
the major mode of news dissemination. Currently, most news is still interpreted by humans but news providers have started to offer newswire products with machine learning
2

Today, Thomson Reuters in one of the major companies for professional financial information and general
professional news dissemination. The company is publicly listed at the Toronto Stock Exchange and
the New York Stock Exchange with 2009 revenues of $12.9 Billion and more than 55,000 employees
worldwide (see 2009 annual report and http://thomsonreuters.com/).

1.2 Structure of the Thesis

3

systems that specifically cater to algorithmic traders. The electron now is not only the basis for news dissemination but it is also in the process of taking over the analysis of news.
Computer systems drastically facilitate the information processing capabilities of market
participants and the speed of information processing.
News has come a long way since Paul Julius Reuter founded Reuters more than 150 years
ago. However, the basic requirements to providers of news, which is relevant for financial
markets, remain the same: accuracy, speed, and impartial distribution. News messages are
one major part of the public information set available to traders and investors in financial
markets. News and information in general have a profound impact on the functioning
of financial markets and price dynamics. Despite news’ long history of importance for
financial markets, the relation of news to financial markets and how this set of public information translates into prices still lacks understanding. This thesis sheds light on the
question how newswire messages, specifically machine-readable newswire messages as one
form of public information, influence modern computerized equity markets. More generally, this thesis studies how markets process information and translate it into security
prices. How information is incorporated into security prices is a key issue in financial
markets research and essential for the understanding of financial markets.

1.2 Structure of the Thesis
The main part of this thesis3 is structured into three chapters. Chapter 2 is based on a joint
working paper with Ryan Riordan and Martin Wagener (cf. Riordan, Storkenmaier, and
Wagener, 2010b). Chapter 3 is based on a working paper with my co-authors Martin Wagener and Christof Weinhardt (cf. Storkenmaier, Wagener, and Weinhardt, 2010) whereas
Chapter 4 is based on a joint working paper with Markus Höchstötter and Ryan Riordan
(cf. Höchstötter, Riordan, and Storkenmaier, 2011). Each chapter focuses on a different
aspect of public information in modern equity markets and on how information is processed through markets. All chapters base on the same news data set as their proxy for
public information. The news data set comprises of Thomson Reuters newswire messages
which additionally feature tags for an automatic processing in empirical analyses based on

3

Financial support by the IME Graduate School at the Karlsruhe Institute of Technology (KIT) funded
by Deutsche Forschungsgemeinschaft (DFG) and financial support by the Karlsruhe House of Young
Scientists (KHYS) is gratefully acknowledged.

1 Introduction

4

machine learning techniques.4 Tags include information on the tone of a news message, its
relevance, and its novelty.
Chapter 2 studies the impact of intraday firm specific public information on intraday
price discovery, liquidity, and trading intensity in a pure electronic limit order market.
The analysis is based on trading at the Toronto Stock Exchange since it operates a pure
electronic limit order book which is same market model operated by many international
exchanges. Additionally, the Canadian equity market has a low degree of fragmentation
of order flow during the observation period. The reaction to news is studied separately
for novel positive, negative, and neutral news messages which arrive during normal trading hours. Most existing empirical research is not able to ex-ante differentiate newswire
messages by their tone (cf. Ranaldo, 2006). Theoretical models suggest varying pre-news
information gathering and post-news information processing capabilities of market participants (Kim and Verrecchia, 1991, 1994). Chapter 2 specifically addresses the following
research questions:
• How do firm specific newswire messages separated into positive, negative, and
neutral messages affect high-frequency price discovery, liquidity, and trading
intensity in an electronic limit order market?
• How do firm specific newswire messages in such a setting impact the highfrequency interaction between price discovery, liquidity, and trading intensity?
Chapter 3 studies the impact of firm specific public information on trading in fragmented markets and particularly on the price discovery process, liquidity, and trading
intensity. The analysis is based on a sample of FTSE 100 stocks traded on the London
Stock Exchange and on Chi-X, the largest multilateral trading facility in Europe. Daily aggregate values for both trading characteristics and newswire messages are used. The impact
of public information on trading is based on a comparison between positive, negative, and
neutral news days. FTSE 100 stocks are suitable for this study since they exhibit a high
degree of fragmentation during the observation period with one large multilateral trading
facility, Chi-X, as the major second trading venue after the LSE. In fragmented markets,
public information has multiple opportunities to translate into prices which leads to the
following research questions:
4

I am grateful to Thomson Reuters for providing access to Thomson Reuters NewsScope Sentiment Engine
archive data.

1.2 Structure of the Thesis

5

• How does positive or negative firm specific news impact the price discovery
process, liquidity, and trading intensity of individual markets in a fragmented
environment?
• How does firm specific news in this setting influence characteristics of market fragmentation and shifts in price discovery, liquidity, and trading intensity
between trading venues?
Chapter 4 shifts the focus from high-frequency, intraday, and daily analyses to international equity markets and comovement measured on a monthly level. It studies the
influence of the flow of firm specific public information on stock market comovement,
thus also on idiosyncratic stock price variability. Existing literature suggests that timevarying characteristics of stock return comovement are influenced by information production (Brockman et al., 2010). The stock return comovement measure in Chapter 4 is
based on a common measure developed by Campbell et al. (2001). I use a direct measure
of firm specific information based on Thomson Reuters newswire messages. Existing literature also suggests that stock return comovement is heavily influenced by a country’s
institutional setting (Morck et al., 2000) which might also influence the association of firm
specific information and stock return comovement. To clarify the influence of firm specific public information on the time-varying characteristics of stock return comovement,
Chapter 4 addresses the following research questions:
• How does the relative firm specific public information flow in an entire market
drive the time-varying component of stock return comovement?
• How do country specific financial development and transparency characteristics influence the association of firm specific public information and stock
return comovement?

Chapter 2
High-Frequency Market Dynamics and
Public Information
2.1 Introduction
Most professional traders observe newswires like Thomson Reuters, Bloomberg, or Dow
Jones. They spend a considerable amount of money on such information sources and
emphasize the importance of speed and accuracy of news. Newswire messages represent
much of the real-time information traders receive. In general, information is central to efficient financial markets and the formation of prices. The intraday high-frequency impact of
newswire messages, especially in today’s automated equity markets, is however still little
understood. It its unclear whether newswire messages contain new information, whether
traders act in advance of or after such messages, and how such newswire messages impact
the price dynamics in modern electronic limit order markets.
This chapter studies the impact of Thomson Reuters newswire messages on the intraday price dynamics of stocks traded at the Toronto Stock Exchange, a modern electronic
exchange. The Toronto Stock Exchange is specifically suitable for such an analysis. First,
it represents a pure limit order book market comparable to most continental European
Exchanges. Second, in contrast to the German or French market, there is no major second language news stream which reduces potential side effects. Third, during this study’s
observation period the Canadian market has a very low level of fragmentation.
To my knowledge this is the first market microstructure analysis to cluster newswire
messages based on content into positive, negative, and neutral messages. The differentiation between positive, negative, and neutral news enables an investigation of asymmet-

2 High-Frequency Market Dynamics and Public Information

8

ric reactions to newswire messages based on sentiment. I find higher adverse selection
costs around news messages. Negative messages induce significantly higher adverse selection costs than positive news messages. Liquidity increases around positive and neutral
messages whereas liquidity slightly decreases around negative messages. Trading intensity
increases around all types of news messages. Summing up, the results suggest different
information gathering and information processing capabilities of market participants and
show asymmetric reactions to good and bad news.
The remainder of this chapter is structured as follows. Section 2.2 introduces related
literature. Section 2.3 gives an overview on the institutional structure of the Toronto Stock
Exchange. Section 2.4 provides a detailed explanation of the used newswire messages, trade
and order book data, and the sample selection process. Section 2.5 introduces the research
design and methodology. Section 2.6 provides results and interpretation and Section 2.7
finally concludes this chapter.

2.2 Related Work
Literature that is related to this chapter can be characterized on two dimensions. The
first dimension is the type of an information event. Information events might be scheduled macroeconomic news, earnings announcements, or relate to media content which
is highly ambiguous and harder to quantify. The Thomson Reuters newswire messages
used in this analysis are somewhat in-between those extremes. They are not as widespread
and ambiguous as arbitrary media content but are rather ambiguous and hard to interpret
in comparison to earnings announcements or macroeconomic news. The second dimension on which related literature can be classified is the temporal scope of the analysis, i.e.
whether an analysis focuses on intraday high-frequency effects or daily impacts. Additionally, literature that relates to information processing in trading is highly relevant for my
analysis. The notion of information in markets, its impact, and its relevance have had
a startling effect on market microstructure research. Since Bagehot (Pseud.) (1971) challenged existing views on the functioning of financial markets and following the seminal
works of Glosten and Milgrom (1985) and Kyle (1985), information has been a central
theme in many market microstructure papers.
The Bagehot (Pseud.) (1971) paper is not a particularly scientific paper, it was published
in a practitioners’ journal under a pseudonym without any empirical or theoretical modelling. It has been later revealed that the author was Jack Treynor, a practitioner in the

2.2 Related Work

9

financial services industry (Treynor, 1995). Just by reasoning without mathematical models Bagehot (Pseud.) (1971) introduces the notion of information asymmetry, the role of
information in trading, and the relation of both to bid-ask spreads. The author postulates
the idea that a spread has to exist without exogenous influences solely based on the fact
that some market participants have private information and that specifically the size of
the spread also depends on information asymmetry, new ideas back in 1971. Although
the article is by no means theoretical or empirical, its basic ideas have inspired much of
information-based market microstructure research in the following years and even decades.
Traditional economics often argue for the irrelavance of the price setting mechanism or
use the fiction of a walrasian auctioneer. However, such assumptions require that trading
is irrelevant for a resulting equilibrium. In situations with asymetrically informed market
participants such assumptions are unlikely to hold and the price setting and information
processing mechanisms matter for the economic outcomes of markets. Glosten and Milgrom (1985) present a theoretical model, formalizing Bagehot (Pseud.) (1971), that explains
how informed market participants reveal information to the market only through trading
and how as a result a bid-ask spread exists purely based on differential information without exogenous transaction costs. Kyle (1985) introduces one of the first models to examine
strategic trading behavior of informed traders. Both theoretical models explain behavior
that can be observed in reality and which could not be explained by previous market microstructure models.
As one of the recent papers closest to my analysis, Ranaldo (2006) analyzes the market
dynamics of firm specific news at the Paris Bourse from an intraday perspective. His six
months news data is based on the Reuters alert system. Additionally, he also analyzes
earnings announcements as a comparison. However, in contrast to my data, he is not
able to differentiate between news messages based on ex-ante news sentiment (i.e. positive,
negative, or neutral). Ranaldo (2006) sorts news data into return bins depending on market
reactions to ex-post differentiate between results but he does not use a measure that is
exogenous to the market. Also my data spans four years instead of only 6 months. He
finds a marginally significant increase in liquidity and slightly lower adverse selection costs
around news arrivals for all types of news. Order books are sufficiently liquid around
news arrivals which shows strong competition for liquidity supply catering an increase in
liquidty demand. Ranaldo (2006) also concludes that “the whole information flow, and
not just earnings announcements, has a significant market impact”. One paper by GroßKlußmann and Hautsch (2011) which uses the same news data that I use, has only appeared

10

2 High-Frequency Market Dynamics and Public Information

when I was in the process of finalizing this thesis. While I focus in this chapter on highfrequency market microstructure effects, Groß-Klußmann and Hautsch (2011) focus on
high-frequency returns, profitability, and as a consequence on the validity of the news data
set. They find that “high-frequency trading activity indeed significantly reacts to intraday
company-specific news items”. Their paper also reveals a general increase of the bid-ask
spread and an increase in trading volume around news messages. However, they do not
differentiate between positive, negative, and neutral news items for market microstructure
changes around news messages.
Krinsky and Lee (1996) analyze the impact of scheduled earnings announcements on
trading at the NYSE and AMEX. They find that the adverse selection component of the
spread increases around earnings announcements while at the same time the order processing and inventory holding costs significantly decline. They attribute this effect to temporary information advantages of informed investors and to faster news processing capabilities of public information processors. The authors use intraday data for their analysis
and cluster a trading day into half-hour intervals. Additional to information asymmetry,
they find an increase in trading volume before and after earnings announcements as well
as an increase in volatility around earnings announcements. In an analysis of the impact
of scheduled macroeconomic announcements on U.S. government bond trading (Green,
2004), results show higher adverse selection costs around macroeconomic news releases
as a consequence of private information impounded through order flow. In contrast to
the Toronto Stock Exchange, US government bond trading is organized as a dealer market
which might yield different results than a public limit order market. Green (2004) controls
for surprise in the empirical model which can be easily done for macroeconomic news
since forecasts are available. In contrast, I cannot control for surprise with firm specific
newswire messages as no forecast or expected value for comparison is available.1 Berry
1

A range of additional studies analyzes the impact of macroeconomic news announcements on financial
markets. Niessen (2007) researches into the effect of media coverage on macroeconomic news processing
in the futures market for government bonds. Her paper provides evidence that there is macroeconomic
information processing prior to economic indicator releases induced through media coverage. Higher
pre-announcement media coverage increases investor attention and leads to stronger post-announcement
market reactions. Evans and Lyons (2008) investigate the effect of macroeconomic news on foreign exchange markets. They analyze a broad spectrum of macroeconomic news and study the direct influence
on prices through order flow. They find that after the announcement of macro economic news there
is more information impounded into the market through order flow than during normal times. This
finding translates into higher adverse selection after macro news than normal and is not consistent with
the hypothesis that public information is directly impounded into the market and directly causes price
changes. Andersen et al. (2007) analyze different futures markets with respect to the release of macroeconomic information. They find quick significant responses also in non-US government bond futures

2.2 Related Work

11

and Howe (1994) also analyze the intraday impact of public information arrivals. Their
proxy for public information is the number of news releases by Thomson Reuters’ news
service per unit of time. They argue that “Reuter’s News is selected as [their] data source
for public information flow because it provides market participants with a timely source
of information on news stories that impact financial markets. [...] Market participants use
this news service on a regular basis, along with Dow Jones News Service and perhaps a few
other newswires, as a prime news source for economic decision making”. This is essentially the same reason why I use Reuters news for the analysis in this chapter. One caveat
of their study is the relatively noisy proxy for public information. Berry and Howe (1994)
only count the number of Reuters news per half-hour interval and additionally focus on
market activity not firm specific trading. Their “results suggest a positive, moderate relationship between public information and trading volume, but an insignificant relationship
with price volatility” (Berry and Howe, 1994).
Little evidence exists on the use of news messages by algorithmic traders. Algorithmic trading is defined as “the use of computer algorithms to automatically make certain
trading decisions, submit orders, and manage those orders after submission” (Hendershott
et al., 2011). Since algorithmic trading data is usually proprietary, researcher often need
to fall back on heuristics. Hendershott and Riordan (2009) present one study that uses
direct proprietary data which enables them to differentiate between algorithmic traders
and human traders. Algorithmic traders in their analysis comprise of both algorithmic
traders implementing human investment decisions and high frequency traders. They find
that algorithmic traders contribute more to price discovery than human traders. Chaboud
et al. (2009) find that algorithmic traders in the foreign exchange market monitor macroeconomic news and pull out of the market for a short time after economic news arrivals
to safeguard themselves against higher adverse selection costs. However, after a short time
of retreat from the market, they provide more liquidity than non-algorithmic traders the
markets. This finding indicates a quick and thus efficient price discovery process. Fleming and Remolona
(1999) find a reduction in trading volume and sharp price reactions with higher volatility after the publication of macroeconomic news in the US Treasury market. In their study, the quoted spread increases
around macroeconomic news announcements and then slowly reverts to normal. They also find highly
significant cross-market linkages between trading venues in different countries. Hess et al. (2008) focus their analysis on the liquidity provision around macrocenomic news announcements in the German
Bund Futures market. They measure liquidity supply through the quoted spread and volume at the best
bid and ask. However, they lack order book data to analyze liquidity at order book levels beyond the best
bid and ask. In their paper, they find that bid and ask volume decreases around macroeconomic news
announcements while the quoted spread increases. This footnote is based on Storkenmaier, Riordan,
Weinhardt, and Studer (2010).

12

2 High-Frequency Market Dynamics and Public Information

hour after macroeconomic news announcements. This result is robust for not only US
nonfarm payroll but also for other US macroeconomic news announcements. Chaboud
et al. (2009) show that at least some news is monitored automatically by computers.
Ederington and Lee (1993) investigate effects of scheduled macroeconomic news on interest rate and foreign exchange futures markets and find higher volatility after news announcements. They study fast reactions with five-minute intervals after the announcement
of macroeconomic information like US nonfarm payroll or the consumer price index. Empirical evidence also suggests that firms assume intraday effects of their announcements and
thus time their information releases (Patell and Wolfson, 1982). Firms release good earnings and dividend announcements intraday and bad ones after trading hours. It seems that
firms which release information assume that the intraday response of trading to good or
bad news matters to the eventual price of their shares. This hints to potential asymmetric
responses of traders to good and bad news.
From a theoretical perspective, Kim and Verrecchia (1991) formulate a model which explains higher adverse selection costs prior to an anticipated announcement such as earnings
announcements. Traders acquire costly private information to trade in advance of a public
announcement. The model also shows what is intuitively clear, the costs for information
gathering influence the magnitude of private information gathering negatively. Higher
marginal costs to aquire information reduce asymmetric information since less private information is gathered pre-announcement. Finally, Kim and Verrecchia (1991) relax the
assumption that an announcement needs to be anticipated. They find that anticipated announcements provide stronger incentives to aquire private information than unanticipated
announcements. Their model “also confirms the intuition that [...] volume arises due to
differential belief revision” (Kim and Verrecchia, 1991). Put into a short sentence, the Kim
and Verrecchia (1991) model shows that pre-announcement information gathering induces
information asymmetry. Kim and Verrecchia (1994) introduce in their model the notion
that different traders have varying capabilities to interpret earnings announcements. The
evaluation of announcements and news depends on a trader’s ability to interpret news but
it might also depend on the support a trader has in analyzing news announcements. In
reality, computers might help to significantly increase the speed of news analysis or staff
that connects announcement information with other information sources enables certain
traders to aquire superior private information from public information sources. Also,
traders have different intellectual capabilities to process information and to process it fast.
All this might lead to an increase in adverse selection costs after earnings announcements

2.2 Related Work

13

due to higher information asymmetry. The Kim and Verrecchia (1994) model predicts
a reduction in liquidity. However, trading volume might still increase despite a decrease
in liquidity around earnings announcements. Another theoretical model developed by
Harris and Raviv (1993) attributes effects around the announcement of public information to speculative trading. Traders disagree as result of differential private information
or different information processing capabilities which leads to a surge in market activity.
The Kandel and Pearson (1995) model is another framework that includes the notion of
differential interpretation of public signals which explains high volumes around public
announcements.
One challenge in the analysis of public information is the transformation of ambiguous
news and media content into variables that can be used in econometric models. Several
papers analyze such ambiguous content and study its impact on financial markets. However, those studies usually base their analyses on daily data, often driven by the nature of
their public information sources. Newspaper content is one of the most frequently studied
types of media content. Niederhoffer (1971) provides one of the earliest papers that analyzes media content. He investigates world events which are defined as having appeared as
a five- to eight column headline in the New York Times. One of his stated objectives is
to “illustrate and suggest specific applications of some techniques for measuring meaning”
(Niederhoffer, 1971). One interesting aspect is that he has the headlines manually classified
by untrained observers based on classifying guidelines; something which is done through
algorithms in more recent research. The paper shows that world events are followed by
larger price changes than normal.
Analyzing Wall Street Journal content seems to be quite popular with financial researchers. Liu et al. (1990) analyze the “Heard-on-the-Street” column in the Wall Street
Journal (WSJ) and find abnormal returns on announcement days in combination with
higher trading volume. The “Heard-on-the-Street” column is a daily column that is supposed to inform readers about developments and news that could potentially have an effect
on stock prices (Liu et al., 1990). The observation period covers more than three years and
comprises of more than 1,000 columns which were all classified manually into buy or sell
recommendations excluding ambiguous columns. The WSJ “Dartboard” column is analyzed by Barber and Loeffler (1993). The column is called “Dartboard” column because
four stocks are randomly chosen by throwing a dart and compared against four stocks recommended by professional investment analysts. Stock market reactions include also positive abnormal returns and higher trading volume. As in the previous study by Liu et al.

14

2 High-Frequency Market Dynamics and Public Information

(1990), Barber and Loeffler (1993) also manually classify the content of the WSJ column.
Tetlock (2007) analyzes the effect of the WSJ column “Abreast of the Market” on the
American stock market and the effect of the market on the column. He finds that high
pessimism in the WSJ column is followed by lower market prices and thereafter by a reversal to fundamentals. He extracts the pessimism factor using computer based content
analysis techniques. The content analysis technique that he applies is based on counting
words that belong to different categories such as positive and negative or active and passive. The automated analysis of content has two advantages. First, manually classifying
4,000 WSJ articles would not be feasible. Second, using a straight forward content analysis
technique does not run the risk to introduce a personal bias in contrast to a manual classification. Tetlock et al. (2008) analyze whether linguistic content comprises information
relevant for financial markets. They find that relevant information that would be hard to
quantify is contained within such content. Their whole paper is focused on quantifying
language in financial news stories. In contrast to Tetlock (2007), they “extend that analysis
to address the impact of negative words in all Wall Street Journal (WSJ) and Dow Jones
News Service (DJNS) stories about individual S&P 500 firms from 1980 to 2004” (Tetlock
et al., 2008). Such an amount of news data would be impossible to classify manually. Tetlock et al. (2008) state that “by quantifying language, researchers can examine and judge
the directional impact of a limitless variety of events, whereas most studies focus on one
particular event type, such as earnings announcements, mergers, or analysts’ recommendations. Analyzing a more complete set of events that affects firms’ fundamental values
allows researchers to identify common patterns in firm responses and market reactions
to events”. This description is comparable to the news data set that I apply in this thesis
which allows for a differentiation between good news and bad news and is available for
the overall firm specific information flow. Tetlock (2008) shows in an analysis of the reaction of investors to stale information about S&P 500 firms that markets react to stale news
through individual overreacting investors but then show subsequent return reversals.
Antweiler and Frank (2004) investigate the link between the information content of Internet stock message boards and financial markets. They use naive bayesian analysis and
support vector machines to classify message board stories. Support vector machines is a
method borrowed from machine learning in computer science where it is often applied to
linguistic content (cf. Joachims, 1998; Tong and Koller, 2001; Leopold and Kindermann,
2002). Antweiler and Frank (2004) find that stock message board postings support the
predicition of market volatility. Disagreement among users of the Internet message boards

2.2 Related Work

15

relates to higher trading volume in line with existing literature that shows that disagreement among traders increases trading activity. One interesting aspect of their work is that
they classify linguistic information which does not follow any styleguides or basic structural principles like newspaper articles and nevertheless still retrieve valuable information.
News media also effects individual buyers’ perception of and their attention towards
specific stocks (Barber and Odean, 2008). Individual buyers are more prone to buy stocks
which have drawn their attention through media outlets because individual investors have
limited resources to consider stock picks. Individual investors can usually choose from
many stocks to buy but mostly sell only stocks which they already have in their portfolios. Traditional theoretical models assume that investors “are equally likely to sell securities with negative signals as they are to buy those with positive signals” (Barber and Odean,
2008), in reality however “for actual investors, the decisions to buy and sell are fundamentally different”. In short, the authors find asymmetric investor behavior which is affected
by news media. Their proxy for news is the Dow Jones News Service. For individual
stocks, Barber and Odean (2008) only discriminate between days with news and days without news. Also the breadth of information dissemination has an influence on stock returns
(Fang and Peress, 2009). Fang and Peress (2009) find that firms without news have higher
returns than firms that are covered by media even when controls are included in the analysis. To study the relation between mass media and returns, Fang and Peress (2009) count
how many articles are published about a specific firm in the New York Times, USA Today,
the Wall Street Journal, and the Washington Post.
All these studies2 have in common that they quantify ambiguous media content or otherwise derive quantitative information from linguistic messages. Analysis techniques range
from only counting news to sophisticated content analysis methodologies. What I study in
this chapter is similar in terms of the content analyzed. The nature of Thomson Reuters
newswire messages is close to newspaper or message board content since it needs some
form of transformation of linguistic messages into variables for econometric analyses. In
my analysis, pre-transformed, already quantified, news messages from Thomson Reuters
are used.
First, this chapter examines how newswire messages separated by their tone affect trading intensity, liquidity, and price discovery. Theory and empirical results suggest that I find
2

There is a range of other studies with daily data concerning public information arrival that provide insight
into information processing and liquidity provision. Among those are Thompson et al. (1987), Fleming
and Remolona (1999), and Ryan and Taffler (2007). Veronesi (1999) provides a theory of asymmetric
influences of good or bad news depending on the prevailing market sentiment.

16

2 High-Frequency Market Dynamics and Public Information

an increase in trading activity around news arrivals. Evidence for liquidity and adverse selection, or more broadly speaking price discovery, is mixed. Financial theory however
suggests an increase in adverse selection costs and a reduction of liquidity. Second, this
chapter investigates how trading activity, liquidity, and adverse selection interact around
news messages. Literature suggests that competition in the limit order book might have an
influence on liquidity supply catering liquidity demand and existing studies also find a rise
in trading activity around news messages (Ranaldo, 2006).

2.3 Institutional Details
The Toronto Stock Exchange (TSX) is Canada’s most important equity exchange operated by the TMX Group.3 The TSX offers trading in equities and equity linked products.
TMX Group also operates the Montréal Exchange which provides futures and derivatives
trading. The historically fragmented Canadian financial exchange sector was reorganized
in 1999. The TSX became Canada’s only senior equities exchange whereas options and
derivates trading was consolidated on the Montréal Exchange. The Alberta and Vancouver exchanges were merged to form the Canadian Venture Exchange4 also operated by
TMX Group5 . In 1997, the TSX closed its floor operations and moved trading to a completely electronic market. The TSX is North America’s third largest equity exchange by
trading volume after Nasdaq and the New York Stock Exchange.6 Canadian exchanges
are regulated on a regional level. The TSX is regulated through the Ontario Securities
Commision7 and through the Investment Industry Regulatory Organization of Canada, a
self-regulatory organization. Prices on the TSX are used to calculate the S&P/TSX 60 index, Canada’s most important stock market index maintained by Standard & Poors. The
index comprises of 60 Canadian incorporated constituents of different industry sectors,
currently representing approximatly 73% of Canadian market capitalization.8
The TSX operates an entirely electronic market with a centralized public limit order
3

Alternative trading systems do not play an important role during the observation period of this analysis.
The TSX’s market share by trading volume was still 94.2% in January 2009, directly after the end of the
observation period, and close to 100% one year earlier. source: Financial Times, 20 November 2009,
“Toronto’s trading platforms draw regulatory scrutiny”.
4
Department of Finance Canada, http://www.fin.gc.ca/toc/2002/cansec_-eng.asp.
5
TMX Group Inc., http://www.tmx.com/.
6
World Federation of Exchanges, 2008, http://www.world-exchanges.org/statistics/.
7
OSC, http://www.osc.gov.on.ca/.
8
Standard & Poors, http://www.standardandpoors.com/.

2.4 Data and Sample Selection

17

book. The market features basic limit and market orders. The TSX market model is based
on price and time priority. Iceberg orders that display only a portion of their total size are
available for a minimum of 500 shares. They sacrifice time priority on the non-displayed
portion of the order. In a centralized limit order book, incoming orders are compared
to existing orders stored in the book. If the price of the incoming order crosses the price
of an existing order, they are matched. The market model also features on-close orders.
Market-on-close orders can be entered until twenty minutes before market closing. Afterwards only contra imbalance side limit-on-close orders are accepted. Order parameters
consist of expiration parameters as well as immediate-or-cancel and fill-or-kill flags. Market
makers, who are essentially liquidity providers, operate within the electronic public limit
order book without proprietary information. Liquidity is solely provided by limit orders
displayed in the order book. TSX market makers are similar to designated sponsors on
Xetra, the electronic limit order market of Deutsche Börse, described by Klar and van den
Bongard (2008).
The TSX’s continuous trading sessions start at 9:30 a.m. and last until 4:00 p.m. local
time equivalent to the New York Stock Exchange. In a pre-opening session, traders can
enter orders but they are not executed until the market opening when continuous trading
starts. Orders cannot be modified at the opening for 20 to 30 seconds before the start of
trading. The TSX co-ordinates the innvocation of circuit breakers that interrupt trading
due to highly volatile markets with US financial markets. However, this chapter focuses
on continuous trading periods.

2.4 Data and Sample Selection
2.4.1 Stock Market Data
Trade and quote as well as order book data are retrieved from the Thomson Reuters DataScope Tick History archive through SIRCA9 . Specifically, I retrieve trade prices and volumes, best bid and ask including associated volumes, and order book data three levels
into the book from 1 January 2005 to 31 December 2008 for S&P/TSX 60 constituents.10
All data entries additionally include Thomson Reuters qualifying codes to identify special trades, quotes, or specific trading sessions. Trades and quotes are timestamped to the
9

Securities Industry Research Centre of Asia-Pacific, I thank SIRCA for providing access to the Thomson
Reuters DataScope Tick History archive, http://www.sirca.org.au/.
10
The analysis also includes a control period with data from 1 January 2003 to 31 December 2006.

18

2 High-Frequency Market Dynamics and Public Information

millisecond. All prices are reported in Canadian dollars. Since the analysis is restricted to
continuous trading, the first and last five minutes of a trading day as well as non-continuous
trading sessions, i.e. curcuit breakers, are removed from the data. This avoids biases associated with the information processing and inventory management processes at those times.
I also delete crossing trades and on-close orders. Thomson Reuters trading data and RNSE
data are timestamped based on the same clock such that timestamps are directly comparable. Tables B.1 and B.2 in Appendix B depict samples of raw trade and quote and raw
depth data for the Toronto Stock Exchange.

2.4.2 News Data
To analyze high-frequency news data, I have access to Thomson Reuters newswire messages. The real-time commercial product is called Thomson Reuters NewsScope Real-time
while I have access to archive data. The Thomson Reuters NewsScope Real-time data
stream is disseminated to approximately 370,000 Reuters screens worldwide. According to
Thomson Reuters, they “deliver over 500,000 alerts and over two million unique stories
a year”11 . These numbers show that Thomson Reuters newswire messages are widely disseminated and read by traders all over the world. Thomson Reuters – also Dow Jones, or
other professional financial news providers – is most probably perceived as more trustworthy by traders and other financial market professionals than rumors on Internet message
boards or television shows. newswire messages provide much of the real-time information
flow available to traders. Thomson Reuters specifically advertises their news streams for
use by algorithmic traders. However, the bulk of newswire readers should still be human.
My data not only comprises normal Thomson Reuters NewsScope Content but is additionally tagged through data generated by the Thomson Reuters NewsScope Sentiment
Engine (RNSE). RNSE allows for a transformation of ambiguous news signals into quantitative computer-readable scores. The Thomson Reuters NewsScope Sentiment Engine
processes news data on three dimensions: sentiment, relevance, and novelty (Thomson
Reuters, 2008a,b). Sentiment reflects the stock specific tone of one news item and is either positive, negative, or neutral. The relevance measure is a stock specific score for a
news item indicating the relevance of a certain news message. Finally, novelty indicates
whether news with the same content has been released prior to a certain news message.
11

Thomson Reuters,
http://thomsonreuters.com/products_services/financial/financial_products/az/newsscope_application_license/.

2.4 Data and Sample Selection

19

News messages are tagged for each stock separately. For example, a news message that is
positive for Google might be negative for Yahoo while it might be much more relevant
for Google than Yahoo. The RNSE analysis is based on machine-learning techniques and
computer linguistics without human interaction. Groß-Klußmann and Hautsch (2011)
also base their analysis on RNSE news data and find that “news engines [in their paper
RNSE] have the potential to successfully pre-structure news”.
Thomson Reuters also provides newswire data that has become an acadamic standard
in the machine learning community for testing text categorization algorithms. Two data
sets are available. One widely used collection of newswire messages is the ‘Reuters-21578’
text categorization test collection which comprises of 21,578 newswire messages from 1987
(cf. Joachims, 1998; Tong and Koller, 2001; Leopold and Kindermann, 2002; Blöhdorn and
Hotho, 2009; Debole and Sebastiani, 2005).12 In the year 2000, this data collection has been
superceded by a new collection of newswire messages called ‘Reuters Corpus, Volume 1’
(RCV1) which includes 810,000 newswire messages from the years 1996 and 1997 (cf. Lewis
et al., 2004; De Melo and Siersdorfer, 2007).13 In addition, multilingual messages called
‘Reuters Corpus, Volume 2’ (RCV2) are also available. The National Institute of Standards
and Technology14 (NIST), a US government research agency, took over the distribution of
RCV1 and RCV2 in 2004.
While I have access to Thomson Reuters NewsScope Sentiment Engine archive data,
professional traders can also purchase the Thomson Reuters NewsScope Sentiment Engine
for real-time news content processing. In contrast to prior research, this unique data set
allows to cluster news based on content and novelty, and also directly associate relevant
news with individual stocks. Table 2.1 reports one sample RNSE news message. An in
depth description of data fields in RNSE data is available in Appendix C.
Newswire data are cleaned by reproducible criteria. First of all, I delete all news that
links to a news message with similar content during the previous twenty-four hours. This
criterium ensures that news messages have a certain novelty and that exactly the same content has not yet been disseminated over Thomson Reuters newswires. Still, I sometimes
find double news in the data which have the same PNAC. PNAC is short for primary
news access code and identifies one story as it develops. This might be a result of technical
irregularities. I keep the first entry and delete all subsequent news messages with the same
12

http://www.daviddlewis.com/resources/testcollections/reuters21578/.
RCV1 and RCV2, http://trec.nist.gov/data/reuters/reuters.html.
14
NIST, http://www.nist.gov/.
13

2 High-Frequency Market Dynamics and Public Information

20

400
350
300
250
200
150
100
50
0

2005

2006

2007

2008

Figure 2.1: Novel Intraday News Per Year and Month on the TSX 2005 to 2008. The figure
shows the number of novel intraday news messages per year and month for the 2005 to 2008 sample.

PNAC within the same day.15 After those initial cleaning procedures, only news messages
that arrive during continuous trading hours on trading days are kept. Overall, I have 6,625
novel intraday news messages for my analysis. Figure 2.1 shows the development of the
number of news over the observation period. The increase in news at the end of 2007
and in 2008 might be a result of the financial crisis during which more newsworthy events
happend than during the previous years. Figure 2.2 shows the distribution of news over
weekdays. Figure 2.3 depicts the number of news for all half-hour intraday periods, also
seperated by news sentiment. The number of neutral news increases sharply during the
last half-hour while before, the overall number of news slightly falls from the beginning of
the trading day. I control in the analysis for potential side effects with time of day dummy
variables and also day of the week dummy variables.
15

Since PNACs are reused by Thomson Reuters’ editorial publishing system, the restriction to the same
PNAC within one day ensures that some completely unrelated news messages are not accidentally
deleted.

2.4 Data and Sample Selection

21

1,600
1,400
1,200
1,000
neutral
negative
positive

800
600
400
200
0

Mon

Tue

Wed

Thu

Fri

Figure 2.2: Novel Intraday News Per Weekday on the TSX 2005 to 2008. The figure shows the
number of novel intraday news messages per day of the week for the 2005 to 2008 sample.

2.4.3 Sample Selection
The firm sample is based on S&P/TSX 60 index constituents from 2005 to 2008. The securities represented in this index are the most actively traded and highest quality publicly
traded Canadian companies and present a broad cross-section of industries. Index constituents are liquid, often and regularly traded, and a considerable portion of a company’s
market capitalization is based on free floating securities. The S&P/TSX 60 currently represents approximately 73% of Canadian equity market capitalization.16 Cleaning the sample,
I additionally require that instruments have to be continuously traded over the years 2005
to 2008.
Then the number of news per company and the number of news per company for each
sentiment category are used to create the sample. To ensure stable estimation results, one
requirement is that the companies in the S&P/TSX 60 index have a minimum amount of
16

Standard & Poors, http://www.standardandpoors.com/.

2 High-Frequency Market Dynamics and Public Information

22

900
800
700
600
500

neutral
negative
positive

400
300
200
100
0

1st 2nd 3rd 4th 5th 6th 7th 8th 9th 10th 11th 12th 13th

Figure 2.3: Novel Intraday News Per Time of Day (half-hours) on the TSX 2005 to 2008. The
figure shows the number of novel intraday news messages for time of day half-hour intervals for
the 2005 to 2008 sample.

overall news messages and a minimum amount of news messages per sentiment category
after all cleaning procedures. Sample companies are require to have at least 80 distinct news
messages over the years 2005 to 2008 and at least 10 news messages per sentiment category
over the years 2005 to 2008 for stable estimation results. As a result, I have a sample of 33
highly liquid, actively traded S&P/TSX 60 constituents. Table 2.2 provides an overview
of the sample and descriptive statistics for each firm in the sample.

2.5 Research Design
2.5.1 Information
To measure information impounded through order flow, I adopt the MRR model introduced by Madhavan, Richardson, and Roomans (1997). I extend the model in the spirit of

2.5 Research Design

23

Green (2004) to include variables for thirty minute intervals around a firm specific news arrival. The MRR model is a common market microstructure model to approximate the adverse selection component of the bid-ask spread, the information revealed through trades.
The authors test their model with data from the New York Stock Exchange. However,
they do not limit their model to markets with specialists or dealers but specifically state
that liquidity providers may also be traders that post limit orders. When informed traders
trade, prices tend to follow their trades which constitutes a risk for market participants
that post limit orders. The adverse selection component measures the part of the bidask spread that is required as a compensation for liquidity suppliers’ risk of losing against
informed traders. Hence, the adverse selection component of the bid-ask spread can be
interpreted as private information that is impounded into prices through the order flow.
The MRR model estimates an adverse selection component as well as an inventory and
order processing cost component of the bid-ask spread.
The MRR model is only based on the trade process and analyzes transaction price
changes and their relation to the trade direction of order flow. I use the standard Lee
and Ready (1991) algorithm to sign trades with contemporaneous quotes as proposed by
Bessembinder (2003a). Bessembinder (2003a) compares different heuristics to infer trade
direction with proprietary data and finds that a comparison of a trade with the respective contemporaneous quote using Lee and Ready’s heuristic provides the best results. In a
limit order market without execution inside the spread, this algorithm signs trades without
ambiguity if trade and quote timestamps match.
Let xi be the trade direction, 1 for a market buy order and -1 for a market sell order, at
time i and pi denotes the transaction price. Specifically, i denotes a single observation in
the trade process. Then, the original Madhavan et al. (1997) model formulates
pi − pi−1 = (φ + θ)xi − (φ + ρθ)xi−1 + εi

(2.1)

as its core concept. Let ρ be the first-order autocorrelation, θ denotes the asymmetric
information component, and φ captures inventory and order processing costs. The original estimation of the model also includes λ in its moment conditions, the probability of
an inside the spread execution which is identified through a trade direction of zero. This
component is not estimated for the TSX. The TSX features a completely electronic limit
order book without inside the the spread executions. The fundamental concept to measure
asymmetric information in this model is that only the deviation from expected order flow

2 High-Frequency Market Dynamics and Public Information

24

comprises information. Expected order flow ρxi−1 is based on order flow autocorrelation
ρ. θ measures “the degree of information asymmetry or the so-called permanent impact of
the order flow innovation” (Madhavan et al., 1997). Beliefs about asset values might change
through new public information without trade and through information impounded by
the order flow whereas the change in belief is positively correlated with order flow innovation. Inventory and order processing costs represent the transitory effect of order flow
on prices. The inventory and order processing cost component of the bid-ask spread is not
dependent on whether a trade is a buy or sell. Bid and ask quotes that reflect the inventory
and order processing cost component are ex-post rational independent of the trade direction. Accordingly, φ is not dependent on the trade process’ autocorrelation ρ. The model
assumes a fixed order size and does not consider trading volume. Madhavan et al. (1997)
address this issue and find that trade direction is better suited to estimate their model than
signed volume.
To analyze information around news, I extend the original MRR model based on Green
(2004). I introduce three dummy variables for thirty minute intervals around a Thomson
Reuters news message arrival and for trading periods not close to newswire messages. One
dummy variable specifies the thirty minutes of trading prior to a Thomson Reuters news
arrival, another dummy variable specifies the thirty minutes after a news arrival, and a
third dummy variable specifies trading periods further away from news releases than thirty
minutes. A dummy variable takes 1 if the observation is within its assigned period around
news releases and 0 otherwise. Let i denote a single observation in the trade point process
and t denotes the distance to a news message in minutes, i.e. t ∈ {−30, −29, . . . , 30}, then
dummy variables are assigned values as follows:
D1,i = 1 if

−30 ≤ t < 0

D2,i = 1 if

0 ≤ t ≤ 30

(2.2)

D3,i = 1 if t > 30 ∨ t < −30
Dummy variables are assigned for news depending on their RNSE news sentiment similar
to dummy variables for intervals around news and no-news periods. One dummy variable
represents positive news, one represents negative news, and one represents neutral news.17
Periods not close to any news are assigned no sentiment dummy variable. Let again i
17

I also test clustering news by relevance. However, price discovery results are essentially the same such that
I do not include the relevance measure into the price discovery and liquidity analysis.

2.5 Research Design

25

denote single observations in the trade point process then sentiment dummy variables are
assigned as follows:
I1,i = 1 if sentiment = +1
I2,i = 1 if sentiment = −1

(2.3)

I3,i = 1 if sentiment = 0
If variables do not take 1 then they take 0. Since I hypothesize that bid-ask spread components change around news announcements also depending on the sentiment of a news
message, the following extended model with dummy variables for different time intervals
and different news types emerges:
pi − pi−1 =

2 X
3 ”
X

—
(φn,m + θn,m )xi Dn,i I m,i − (φn,m + ρn,m θn,m )xi−1 Dn,i−1 I m,i−1 +

n=1 m=1

(φ3 + θ3 )xi D3,i − (φ3 + ρ3 θ3 )xi−1 D3,i −1 + εi
(2.4)
Madhavan et al. (1997) use absolute price changes to estimate the model. To support the
interpretation of results, I use relative price changes in basis points to estimate Equation 2.4
excluding overnight returns. However, results are robust to using absolute or relative price
changes. The model is estimated using the generalized methods of moments (GMM) with
the Newey and West (1987) procedure as proposed in the original paper by Madhavan et al.
(1997). The Newey-West procedure is robust to autocorrelation and heteroskedasticity.
I apply the Newey-West estimator with five lags. Using more lags does not change the
significance of the results. Let ri denote the relative price change which is calculated as


ri = 10, 000 × ln pi / pi−1 then the model can be estimated as
ui = ri +

2 X
3 ”
X

—
−(φn,m + θn,m )xi Dn,i I m,i + (φn,m + ρn,m θn,m )xi−1 Dn,i−1 I m,i−1 −

n=1 m=1

(φ3 + θ3 )xi D3,i + (φ3 + ρ3 θ3 )xi−1 D3,i−1 −

12
X
t d =1

τt d Tt d −

4
X

ωw d Ww d

w d =1

(2.5)
including half-hour dummy variables T t d for the time of day and dummy variables for the
day of the week Wwd . Green (2004) also includes external variables in his analysis, however,
to account for the surprise in macroeconomic news. Excluding the control variables T t d

2 High-Frequency Market Dynamics and Public Information

26

and Wwd from the analysis does not significantly change the results. The constant α and
the parameter vector
β = (θ1,1 , θ1,2 , θ1,3 , θ2,1 , θ2,2 , θ2,3 , θ3 ,
φ1,1 , φ1,2 , φ1,3 , φ2,1 , φ2,2 , φ2,3 , φ3 ,
ρ1,1 , ρ1,2 , ρ1,3 , ρ2,1 , ρ2,2 , ρ2,3 , ρ3 )
are exactly identified by the ordinary least squares (OLS) normal conditions and the following additional moment conditions:


€
Š2
ρ
x
x
D
I
−
x
D
I
 i i−1 1,i−1 1,i −1 € i−1 1,i−1 1,i −1 Š2 1,1 


 xi xi−1 D1,i−1 I2,i −1 − xi−1 D1,i−1 I2,i −1 ρ1,2 


€
Š2
x x D

 i i−1 1,i−1 I3,i −1 − xi−1 D1,i−1 I3,i −1 ρ1,3 
€
Š2



E
 xi xi−1 D2,i−1 I1,i −1 − € xi−1 D2,i−1 I1,i −1 Š ρ2,1  = 0


2
 xi xi−1 D2,i−1 I2,i −1 − xi−1 D2,i−1 I2,i −1 ρ2,2 


€
Š2


 xi xi−1 D2,i−1 I3,i −1 − xi−1 D2,i−1 I3,i −1 ρ2,3 


€
Š2
xi xi−1 D3,i −1 − xi−1 D3,i−1 ρ3

(2.6)

The moment conditions in Equation 2.6 represent autocorrelations of the trade direction
indicator. The estimation of Equation 2.5 provides results for asymmetric information θ,
the inventory and order processing cost or cost of supplying liquidity φ, and the autocorrelation of order flow ρ for each single interval.
To assess the model’s statistical significance, likelihood ratio tests as in Green (2004) are
applied. These tests compare the GMM criterion function of the unrestricted model with
restricted models. Here, the restricted models posit that only one coefficient is needed
for each model to capture adverse selection and costs of supplying liquidity without the
option for those measures to vary around news arrivals. To consider robustness, I also
compare model implied spreads with actual quoted spreads from the data. Since model
implied spreads are solely based on the order flow they do not necessarily need to be exactly
the same as data based quoted spreads. However, they should be roughly similar in their
order of magnitude. The medians of the differences of individual model coefficients of the
thirthy-three stock specific models are compared with Wilcoxon signed rank tests.

2.5 Research Design

27

2.5.2 Trading Intensity, Liquidity, and Volatility
To measure trading intensity, I transform the trade process into a process with one observation per minute and calculate the number of trades per minute, number of shares
traded per minute, and traded dollar volume per minute. For estimation purposes, the
natural logarithms of the number of shares traded per minute and traded dollar volume
per minute are used. For the news dummy variable definition, I resort to the MRR information model definition of dummy variables and use exactly the same. The no-news
dummy variable does not need to be included, it is the basis of comparison and coefficients
capture the difference to no-news periods. I include time dummy variables to account for
market trends. Time dummy variables γq y are included for each quarter for a specific year
into all regressions. In this chapter’s data, 16 year quarter combinations are found. I also
include firm dummy variables F x with x ∈ {1, 2, . . . , 33} where x denotes a single firm. In
the regression model, one firm serves as the base category which results in 32 firm dummy
variables. Additionally, the equation includes half-hour dummy variables T t d for the time
of day and dummy variables for the day of the week Ww d . Let l denote the minutes in the
data and t m x,l denotes the respective trading intensity measure on a minute and per firm
basis then the following model is used to assess trading intensity around news messages:
t m x,l = a +

2 X
3
X

ψn,m Dn,x,l I m,x,l +

n=1 m=1

32
X

ιx Fx +

x=1
12
X
t d =1

τt d Tt d +

15
X

ζq y γq y +

q y=1
4
X

(2.7)

ωw d Ww d + e x,l

w d =1

To estimate the linear model in Equation 2.7, Newey and West (1987) standard errors based
on five lags are used.18
Quote based, ex-ante observable, liquidity is measured based on three different indicators: quoted half spread, the volume at the best bid and ask, and the volume at three depth
levels. All three liquidity measures are based on a quote-to-quote process which is then
aggregated to minute averages for estimation purposes. Let ai denote the best ask and bi
the best bid at time i, then quoted half spreads q si based on Bessembinder and Kaufman

18

Compare Cai et al. (2004) who also use the Newey and West (1987) standard errors for an intraday analysis.

2 High-Frequency Market Dynamics and Public Information

28

(1997) are calculated as follows in basis points:
q si =

‚

ai − bi

Œ

(ai + bi )/2

/2 × 10, 000

(2.8)

Then quoted spreads q si are aggregated to per firm and minute average quoted spreads
q s x,l . Quoted spreads are also calculated as trade-time quoted spreads for which I need the
trade process. Those quoted spreads capture liquidity represented through the best bid and
ask at the time of trades. Quoted spreads, however, only capture liquidity independent of
trade size. To further analyze liquidity, I consider Canadian dollar volume at the best bid
and ask (Depth0). Let again be ai the best ask, bi the best bid, ani the number of shares
available at the ask, and b ni the number of shares available at the bid then the Canadian
dollar volume at the best bid and ask e vi is calculated as
e vi = b ni × bi + ani × ai .

(2.9)

Then Depth0 e vi is aggregated to per firm and minute average Depth0 e v x,l . For estimation purposes, the natural logarithm of Canadian dollar volume at the best bid and ask is
used (l e v x,l = ln e v x,l ). Depth0 only provides information about volume directly at the
spread. Order book data allows to analyze available volume deeper into the book and liquidity which is used and needed for larger trades. In combination with the other liquidity
measures, depth at three levels (Depth3) allows for a much more precise analysis of liquidity than quoted spreads and bid-ask volume alone. Let ai,d l be the ask at time i on depth
level d l , bi,d l denotes the bid on depth level d l , ani,d l is the Canadian dollar volume on a
certain depth level at the ask, and b ni,d l denotes the volume at depth level d l on the bid.
Then the depth measure di for three depth levels is calculated as
di =

3
X
d l =1

b ni,d l × bi,d l +

3
X

ani,d l × ai,d l .

(2.10)

d l =1

Again, Depth3 di is aggregated to per firm and minute average Depth3 d x,l . As for volume
at the best bid and ask the natural logarithm of Depth3 is used for the estmation (l d x,l =
ln d x,l ).
The effective spread, a trade process based liquidity measure, is the spread paid when
a market order is executed against a limit order in the order book and as a trade based

2.5 Research Design

29

measure takes into account available depth. The effective spread also captures institutional
features of a market such as iceberg orders. Let pi be the execution price and Di the trade
direction then the effective spread e si is defined as
e si = Di ×

pi − (ai + bi )/2
(ai + bi )/2

× 10, 000.

(2.11)

The model to estimate the impact on liquidity measures is comparable to the one for trading intensity (Equation 2.7). Before I estimate the models, all liquidity measures are aggregated to minute data to have approximately the same number of observations as for trading
intensity. Let l be the indicator for one minute, x for a firm, and l m x,l the liquidity measure under consideration for a firm and minute. Using the same definition for dummy
variables as for trading intensity, the following model emerges:
l m x,l = a +

2 X
3
X

ψn,m Dn,x,l I m,x,l +

n=1 m=1

32
X

ιx Fx +

x=1
12
X

τt d Tt d +

t d =1

15
X

ζq y γq y +

q y=1
4
X

(2.12)

ωw d Ww d + e x,l

w d =1

I estimate the model exactly like the one for trading intensity with Newey and West (1987)
standard errors.
To estimate realized volatility, also known as realized variance (cf. Hansen and Lunde,
2005), I construct one minute midpoint to midpoint returns from quote data. With realized volatility, I assess high-frequency volatility changes around newswire messages depending on the news sentiment (negative, positive, or neutral). To calculate the square of
returns for the realized volatility, logarithmic midpoint returns are used. Let m p l denote
a one minute midpoint then the one minute realized volatility is defined as
r vl =

‚
ln

m pl
m p l −1

Œ2
× 10, 000.

(2.13)

The original realized volatility measure is multiplied by 10,000 to enhance readability of
the numbers.19 Scaling realized volatility by 10,000 does not change its statistical properties. The same regressions as above are used to analyze realized volatility around newswire
19

Usually, one minute returns are quite small in magnitude which would lead to very small numbers in the
result tables.

30

2 High-Frequency Market Dynamics and Public Information

messages for news with different sentiments. All measures are winsorized at 0.1% and
99.9% to account for potential extreme values through technical data recording errors.

2.5.3 Returns and Profitability
To assess basic profitability, I calculate excess returns for different intervals g around a
news announcement. Returns are calculated from thirty minutes before news arrivals up
to the news arrival, from a news arrival to thirty minutes after a news arrival, and from
thirty minutes before to thirty minutes after news arrive through the Thomson Reuters
system. Ten minute returns around news arrivals are calculated equivalently to returns
thirty minutes around news arrivals. Let r x,g be the simple stock specific return, v g denotes
the TSX/S&P 60 return over the same time intervall, p x,g denotes a stock specific price
whereas p gs denotes the index price then excess returns are defined as
z x,g = r x,g − v g = ln

p x, g
p x,g −1

− ln

p gs
p gs −1

.

(2.14)

Let S x,g denote sentiment for a return and W x, g denotes relevance. I regress returns on
sentiment multiplied by relevance since I hypothesize that, for profits, news with higher
relevance should somehow correlate with higher price jumps. Then the following regression with firm dummy variables emerges:
z x,g = a + f (S x,g × W x,g ) +

32
X

F x + e x,g

(2.15)

x=1

I perform regressions on all news and on groups of news each missing either positive,
negative, or neutral news. Standard errors are White (1980) heteroskedasticity consistent
standard errors. Durbin-Watson tests show little autocorrelation in the residuals of regressions with news based returns as the dependent variable, which is something one would
expect given the fact that there is no certain time interval between different unscheduled
news.

2.6 Results and Interpretation

31

2.6 Results and Interpretation
For each stock in this chapter’s data set, I collect information on the number of news items
and average sentiment. The descriptive statistics to the sample are contained in Table 2.2.
The average firm has 201 distinct news items, and an average sentiment of -0.0356 which
is marginally negative. A firm specific sentiment of 0 for a single news item indicates that
sentiment for this news item in combination with the specific firm is neutral. The average
sentiment is in line with studies that report a bias of news media to report more on negative
than on positive events (Soroka, 2006). In total, I observe 6,625 information events derived
from firm specific news messages over a period of four years. The average firm market
capitalization over the years 2005 to 2008 is approximatly C$24bn. Market capitalization
ranges from C$5.5bn to C$63bn with an accumulated market capitalization of C$791bn.
The overall domestic market capitalization of firms traded on the TSX was C$1,256bn at
the end of 2008, comparable to the market capitalization at Deutsche Börse in Germany.20
This is an indication that the firm sample comprises a large share of Canadian market
capitalization. Table 2.2 also shows that the sample represents a broad cross-section of
industries.
The focus of the analysis lies on intraday price dynamics and I calculate a number of appropriate descriptive measures. In Table 2.3, summary statistics are presented for periods
without news, before and after news for positive, negative, and neutral news separately.
Descriptives in this table are not yet adjusted for the year and quarter and for firm specific
effects. I present results for each ‘setting’ for quoted spreads over all quote changes and for
quoted spreads only at trade-time, effective spreads, volume at the best (Depth0), depth at
three levels into the order book (Depth3), trading intensity (number of trades per minute),
numbers of shares traded per minute, volume per minute, and share price volatility (realized volatility).

20

World Federation of Exchanges, http://www.world-exchanges.org/statistics/.

2 High-Frequency Market Dynamics and Public Information

32

Table 2.1: Sample News. Table 2.1 shows one novel intraday RNSE news message for the firm
‘Research in Motion’ (RIM.TO).
Sample RNSE News Item - TSX
timestamp
bcast_ref
stock_ric
item_id
relevance
sentiment
sent_pos
sent_neut
sent_neg
lnkd_cnt1
lnkd_cnt2
lnkd_cnt3
lnkd_cnt4
lnkd_cnt5
lnkd_id1
lnkd_id2
lnkd_id3
lnkd_id4
lnkd_id5
lnkd_idpv1
lnkd_idpv2
lnkd_idpv3
lnkd_idpv4
lnkd_idpv5
item_type
item_genre
bcast_text
dsply_name
pnac
story_type
cross_ref
proc_date
take_time
story_date
story_time
named_item
take_seqno
attribtn
prod_code
topic_code
co_ids
lang_ind

24 OCT 2007 16:30:02.064
RIM.TO
RIM.TO
2007-10-24_16.30.01.nN24487523.T1.8da5a8b6
0.150756
1
0.559651
0.358283
0.0820664
0
0
0
0
0
.
.
.
.
.
.
.
.
.
.
ARTICLE
NOT DEFINED
RIM rolls out Facebook software for BlackBerry
2
nN24487523
S
.
24-OCT-2007
16:30:01
24-OCT-2007
16:30:01
.
1
RTRS
E U CAN G PSC RNP DNP PGE PCO PCU EMK
BUS CA US DE INV TEL WWW SFWR HDWR ENT LEI TEEQ
TECH COMS ELC CEEU EUROPE WEU LEN RTRS
RIMM.O RIM.TO DT.N
EN

2.6 Results and Interpretation

33

Table 2.2: Descriptive Statistics for Sample Companies. The sample is based on stocks continuously listed in the S&P/TSX 60 index between 2005 and 2008. 33 stocks qualify for the sample after
filtering based on newswire data. Table 2.2 reports descriptive statistics for the number of news,
news sentiment and relevance, market value, and economic sector. News measures are derived from
Thomson Reuters RNSE data whereas market value and economic sector are based on Compustat
data. The average per company sentiment is denoted ‘Sent’. The overall number of news (#) as
well as the number of news differentiated by sentiment are reported. ‘MVal’ stands for the average
market value in Million Candian dollars. The table is sorted by the descending number of news per
company in the analysis.
Company Name

#News

#+

#-

#o

Sent

MVal

Economic Sector

Barrick Gold
Research in Motion
Royal Bank of Canada
EnCana
Toronto-Dominion Bank
Nortel Networks
Bank of Nova Scotia
Goldcorp
Canadian Imperial Bank
BCE
Petro-Canada
Bank of Montreal
Suncor Energy
Cameco
Potash Corp. of Sask.
Can. Natural Resources
Can. National Railway
Bombardier
Teck Resources
Imperial Oil
Enbridge
Telus
TransCanada
Agrium
Kinross Gold
Nexen
Talisman Energy
Manulife Financial
National Bank of Canada
Magna International
Rogers Communications
Yamana Gold
Agnico-Eagle Mines

518
366
361
302
290
287
270
252
247
239
215
205
205
201
193
179
178
175
175
162
158
145
143
139
134
133
131
118
118
114
96
91
87

156
112
125
102
103
86
119
68
59
92
74
70
68
87
69
59
52
69
70
57
51
52
87
53
30
37
44
43
40
40
38
29
26

169
181
134
129
135
116
83
67
135
100
75
84
89
41
89
83
94
58
67
54
69
68
39
46
39
52
36
44
52
50
38
13
37

193
71
102
71
52
85
68
117
53
47
66
51
48
73
35
37
32
48
38
51
38
25
17
40
65
44
51
31
26
24
20
49
24

-0.0251
-0.1896
-0.0249
-0.0894
-0.1103
-0.1045
0.1333
0.0040
-0.3077
-0.0335
-0.0047
-0.0683
-0.1024
0.2289
-0.1036
-0.1341
-0.2360
0.0629
0.0171
0.0185
-0.1139
-0.1103
0.3357
0.0504
-0.0672
-0.1128
0.0611
-0.0085
-0.1017
-0.0877
0.0000
0.1758
-0.1264

30,944
32,982
62,805
45,014
45,919
8,922
45,957
21,011
27,133
25,807
21,602
29,202
36,976
21,713
24,745
32,472
23,564
7,030
12,575
40,966
14,086
15,428
20,011
6,510
8,689
14,857
18,680
52,332
8,876
8,233
22,898
5,573
6,385

Materials
Info. Tech.
Financials
Energy
Financials
Info. Tech.
Financials
Materials
Financials
TelCo Services
Energy
Financials
Energy
Energy
Materials
Energy
Industrials
Industrials
Materials
Energy
Energy
TelCo Services
Energy
Materials
Materials
Energy
Energy
Financials
Financials
Consumer Discr.
TelCo Services
Materials
Materials

Mean
Standard Deviation

201
93

69
31

78
41

54
34

-0.0356
0.1278

23,967
14,965

Median

178

68

68

48

-0.0672

21,620

Minimum
Maximum

87
518

26
156

13
181

17
193

-0.3077
0.3357

5,573
62,805

6,625

2,267

2,566

1,792

Sum

790,914

4.7422
4.7786

4.7480
4.7808

4.4162
4.0826

4.3673
3.9593

4.7265
5.5519

4.6377
5.5925

4.5722
4.1409

4.4957
3.7754

Overall
Mean
StdDev

No news
Mean
StdDev

Before News (positive)
Mean
StdDev

After News (positive)
Mean
StdDev

Before News (negative)
Mean
StdDev

After News (negative)
Mean
StdDev

Before News (neutral)
Mean
StdDev

After News (neutral)
Mean
StdDev

QSpread
(in bps)

3.6117
3.5745

3.6538
3.9847

3.7116
5.5496

3.7557
5.4426

3.4886
3.7472

3.5294
3.8690

3.7681
4.3926

3.7639
4.3977

QSpreadT
(in bps)

3.6653
3.5820

3.7091
3.9928

3.7622
5.5166

3.8097
5.4032

3.5362
3.7556

3.5746
3.8552

3.8159
4.3960

3.8118
4.4004

ESpread
(in bps)

148
134

145
127

126
122

121
115

143
135

142
143

132
116

132
116

Depth0
(in C$1,000)

476
516

476
537

411
522

399
465

461
488

458
523

415
401

416
404

Depth3
(in C$1,000)

16.08
18.90

15.57
17.76

15.97
17.89

16.93
19.16

13.16
15.73

13.55
15.99

9.53
11.73

9.70
11.97

#TradesMin

9,503
21,084

8,744
22,770

8,090
21,126

8,343
21,407

7,906
21,781

8,188
23,887

5,667
17,025

5,746
17,197

#SharesMin

373
487

330
441

341
484

357
490

335
516

343
498

230
345

234
352

VolumeMin
(in C$1,000)

0.01517
0.07905

0.01759
0.11895

0.01628
0.09143

0.02277
0.14118

0.01117
0.05413

0.01710
0.13380

0.01183
0.11035

0.01196
0.11027

RV

Table 2.3: Descriptive Statistics Market Measures. Table 2.3 provides descriptive statistics for the market measures over all companies in the
sample. Descriptives are shown overall and for different news periods and no-news periods. ‘QSpread’ denotes the average quoted spread per
minute whereas ‘QSpreadT’ denotes the average quoted spread per minute at trades. ‘ESpread’ denotes the average effective spread per minute.
‘Depth0’ is the average Canadian dollar volume at the best bid and ask and ‘Depth3’ is the dollar volume three levels into the order book.
‘#TradesMin’ is the average number of trades per minute, ‘#SharesMin’ the average numbers of shares traded per minute and ‘VolumeMin’ the
average Canadian dollar volume traded per minute. ‘RV’ represents realized volatility based on minute-to-minute midpoint returns. Spread
measures are in basis points (bps). Measures are calculated for the years 2005 to 2008 over the whole sample.

34
2 High-Frequency Market Dynamics and Public Information

2.6 Results and Interpretation

35

The overall average quoted spread is 4.7422 basis points (bps), the average quoted spread
at trade-time is 3.7639 bps, and the average effective spread is 3.8118 bps. Spreads on the
TSX are very small, thus they are evidence for a generally highly liquid market. Since the
TSX does neither feature hidden liquidity inside the spread nor inside the spread executions, effective spreads are on average slightly smaller than quoted spreads at trade-time.
However, general quoted spreads are significantly larger than effective spreads which indicates that market monitoring occurs and market participants trade when it is comparably cheap to trade. Market participants are able to monitor quoted spreads, Depth0, and
Depth3 ex-ante. Effective spreads and exact timestamps for quoted spreads at trade-time
can only be observed ex-post. The small difference between quoted spreads at trade-time
and effective spreads indicates that there is often sufficient liquidity at Depth0, even during news periods. Depth3 (avg. C$416k) is approximately three times higher than Depth0
(avg. C$132k) which shows in combination with quoted spreads at trade-time and effective
spreads that sufficient liquidity exists on average on the first three levels of the order book.
Trading intensity measures also show that the sample is actively traded with an average of
ten trades per minute and firm. Descriptive statistics provide some evidence that liquidity
increases around positive and neutral news. The average effective spread during no news
periods is 3.8159 bps, for positive news it is slightly lower before news with 3.5746 bps
and 3.5362 bps after news. Effective spreads before neutral news are on average 3.7091 bps
and they are 3.6653 bps after neutral news. The values are inconclusive for negative news.
In line with existing literature, I find that each of the measures of trading activity is larger
around news announments (cf. Green, 2004; Berry and Howe, 1994; Liu et al., 1990). To
further investigate and estimate information, liquidity, trading intensity, and volatility, the
regression models presented in Section 2.5 are applied.

Median Est.
Mean Est.
Median t-stat
Median Est.
Mean Est.
Median t-stat

Order Proc.
φ

Autocorr.
ρ

continued on next page . . .

Median Est.
Mean Est.
Median t-stat

Adv. Selection
θ

0.3500
0.3585
(565.91)

1.1374
1.5922
(198.81)

1.4741
1.4282
(270.50)

no news (nn)

0.3948
0.4326
(46.68)

1.0011
1.3123
(17.37)

1.7630
1.6808
(25.71)

before news

0.3996
0.4443
(54.75)

0.8889
1.2335
(13.37)

1.5632
1.6478
(26.20)

after news

positive

0.4176
0.4229
(60.01)

1.0131
1.5121
(16.72)

1.5574
1.8309
(28.88)

before news

0.4050
0.4401
(66.81)

0.8777
1.1915
(13.19)

1.6086
1.9807
(28.09)

after news

negative

Panel A: By-Company Information Estimations - MRR Model

0.4270
0.4646
(50.65)

0.7810
1.1322
(11.95)

1.6194
1.8085
(23.81)

before news

0.3998
0.4231
(42.57)

1.0036
1.2462
(16.81)

1.3884
1.7461
(20.65)

after news

neutral

Table 2.4: Information Estimations Around News. Table 2.4 provides the results of the MRR model for no-news and news periods. Results
comprise of the adverse selection components θ, order processing costs φ, and trade autocorrelations ρ. The terms positive, negative, and
neutral relate to the RNSE news sentiment. ‘Before news’ and ‘after news’ describe thirty minute intervals before and thirty minute intervals
after a news message is disseminated over Thomson Reuters’ news wire systems. The MMR model is estimated on a per company basis for the
years 2005 to 2008. By-company estimation results in Panel A consist of the medians and means of GMM estimation results for each single
company in the sample. Robust median t-statistics can be found below estimates in parantheses. Panel B provides differences between different
intervals and no-news periods and between pre- and post-news periods. The medians of the differences ∆Est are compared with Wilcoxcon
Signed Rank tests. ‘a’ denotes significance at the 0.1% level, ‘b’ at the 1% level, and ‘c’ at the 5% level.

36
2 High-Frequency Market Dynamics and Public Information

-0.0606c
0.0116
0.0432 b
0.0015

Median ∆Est
p-value
(Wilcoxon T.)

Median ∆Est
p-value
(Wilcoxon T.)

Autocorr.
ρ

0.0594a
0.0004

-0.1524a
0.0007

0.0098

0.0015

Order Proc.
φ

0.0861 b

0.1524 b

Median ∆Est
p-value
(Wilcoxon T.)

after - nn

before - nn

positive

0.0312a
0.0002

0.0298
0.4888

< .0001

0.2000a

before - nn

0.0440a
0.0006

-0.1215c
0.0445

< .0001

0.2140a

after - nn

negative

0.0690a
< .0001

-0.2142a
< .0001

0.0043

0.1965 b

before - nn

0.0289c
0.0192

-0.0669c
0.0262

0.0368
0.3655

after - nn

neutral

Panel B: By-Company Information Estimation Differences - MRR Model

Adv. Selection
θ

. . . continued from Table 2.4

-0.0204
0.4778

-0.0081
0.5576

0.0649
0.5815

before - after

positive

0.0023
0.7071

0.1155
0.1754

0.0178
0.6812

before - after

negative

0.0217
0.2947

-0.1015
0.0980

0.0749
0.1813

before - after

neutral

2.6 Results and Interpretation
37

2 High-Frequency Market Dynamics and Public Information

38

2.6.1 Information
To understand information processing around news announcements, I present the results
of the extended MRR model. Theory (cf. Kim and Verrecchia, 1994) suggests that informed trading after information events should increase. Depending on the type of an
event, scheduled versus unscheduled, and the instrument traded, informed trading may
also increase before an event (cf. Kim and Verrecchia, 1991). Differential pre-news announcement information gathering capabilities of market participants and varying postnews information processing capabilities lead to differential news interpretation.
In general, the MRR estimation is quite consistent with quoted and effective spreads calculated from the trade and quote data. Since spread components are estimated based on the
trade point process in the MRR model, they do not necessarily need to exactly compare to
quoted spreads. However, they should be comparable in magnitude. Panel A of Table 2.4
provides the estimates for the adverse selection component of the spread, the order processing cost component, and the trade autocorrelations. The sum of adverse selection costs
and order processing costs during no-news periods based on the mean estimate is 2.6115
bps (1.4741 + 1.1374) which is not too far off the quoted spread at trade-time of 3.7639
bps. The adverse selection component of the spread is 1.4741 bps during no-news periods
and increases for all settings around news announcements except after neutral news. I find
consistent with intuition, and in contrast to Green (2004) who analyzes a dealer market,
a positive order processing cost component of spreads for all news settings. Order processing costs fall as adverse selection costs increase. In comparsion, adverse selection is the
larger of the two spread components.
Median t-statistics of all MRR coefficients are highly significant. Table 2.7 provides
statistics of the likelihood ratio tests for the MRR models. I report mean and median
χ 2 statistics as well as the number of significant individual models at the 0.1% level out of
the 33 sample firms. The χ 2 statistics for the likelihood tests for adverse selection and order processing costs are highly significant with median χ 2 values of 194 and 189. I find that
out of 33 individual models, 32 models are highly significant at the 0.1% level for adverse
selection costs and 32 are also highly significant for order processing costs.
Panel B of Table 2.4 provides information on the difference between no-news periods
and news periods before and after news arrivals for positive, negative, and neutral news.
As one would expect, adverse selection costs are higher after positive and negative news
than in periods without news. The only period around news messages in which adverse

2.6 Results and Interpretation

39

selection is not statistically different to no-news periods, is after neutral news. The differences for positive news of 0.1524 bps and 0.0861 bps are significant at the 1% level. The
values for negative news at 0.2000 bps and 0.2140 bps are higher than the values for positive
news and also highly significant at the 0.1% level. Neutral news messages show a significant increase in adverse selection pre-news arrival and no statistically significant effect after
arrival in comparison to no-news periods.
Interesting is the highly significant increase in adverse selection around negative news in
comparison to positive news. As described above, the theoretical models of Kim and Verrecchia (1991, 1994) predict higher adverse selection costs around information events. Market participants put different levels of effort into information gathering. Consequently,
some market participants are better informed than others pre-announcement. Differential
levels of private information raise the level of information asymmetry in a market prenews which induces higher adverse selection costs. In general, it does not matter how the
pre-announcement information is acquired. It might be that this information is driven
by insider trading (information leakage) or it could be more innocuous such as news announcements before the Reuters’ release by a competitor or other information sources,
e.g. rumors. Krinsky and Lee (1996) provide empirical evidence for Kim and Verrecchia
(1991, 1994) and find higher adverse selection costs around announcements comparable to
my results for positive and negative news. Before neutral news arrive at a market, adverse
selection costs are higher than normal but although being higher post-news arrival the
difference is not statistically significant. One possible explanation in the light of existing
models is that information gatherers cannot agree pre-news whether information is positive or negative which induces higher adverse selection costs. What exactly happens post
neutral news arrivals is however not entirely clear.
The main differences between my study and those of others are that Thomson Reuters’
firm specific newswire messages are generally unscheduled and that I have an ex-ante exogenous tone (positive, negative, or neutral) for news messages. Ranaldo (2006) attempts
to solve this problem ex-post by sorting events into return bins after the arrival of a news
message. However, my data is potentially better suited to reflect traders’ impression of
news. It is interesting that there are significant differences between positive, negative, and
neutral news and the positive and negative ones in particular. Traditional finance theory
does not differentiate between positive and negative public information. However, psychological studies from the field of impression formation show that humans react stronger
to bad news than to good news, they react asymmetrically (Soroka, 2006; Ronis and Lip-

2 High-Frequency Market Dynamics and Public Information

40

inski, 1985). As one example from the finance literature, Tetlock (2007) only finds significant market reactions to bad news in a Wall Street Journal column. There is additional
finance literature that finds asymmetric reactions of market participants to good and bad
information in general. Akhtar et al. (2011) study the effect that the monthly release of the
Australian consumer sentiment has on the Australian stock market. They find a significant
impact of bad information while good information does not have a statistically significant
effect on the stock market. They attribute this result to the ‘negativity effect’ found in
psychology literature. Stock markets react stronger to monetary policy decisions that are
bad for stock markets than to those that are good for stock markets on an intraday level
(Chuliá et al., 2010). Chen et al. (2003) find that “negative news from the US market will
cause a larger decline in a national stock return [i.e. non-US market returns] than an equal
magnitude of good news”.
A concept that provides additional insights into the different results for positive and negative news may be found in the literature on ambiguity and ambiguity aversion.21 People,
and as such also market participants, do not like ambiguity and prefer known over unknown risk. Ambiguity aversion has been further developed within financial models (cf.
Epstein and Schneider, 2008, 2010; Leippold et al., 2008; Gagliardini et al., 2009). Text can
generally be classified as ambigous information, in that the information content is more
difficult to interpret than the price signals generated in markets (trades and quotes). Ambiguity averse traders react asymmetrically to ambiguous information (Epstein and Schneider, 2008), if the information is positive they act as if they are unsure of the precision of
the information, and if information is negative they act as if it is precise. If the market
is composed of a proportion of investors that exhibit ambiguity aversion, this may help
to explain the fact that the adverse selection costs around negative news are higher than
around positive news. Both, the more psychological view of asymmetric reaction and the
more economics oriented model of ambiguity aversion, base on the same understanding
of human nature and provide explanations to observed trader behavior in this chapter.
21

The fundamental concept of ambiguity aversion is based on a hypothetical experiment by Ellsberg (1961).
In an experiment, there are two boxes with red and blue balls. Both boxes contain 100 balls. One
box contains 50 blue and 50 red balls whereas the distribution of balls in the other box is unknown to
participants. Subjects are now asked to draw a ball. They play the game twice and receive a payoff of
e.g. 100 if they draw a red ball the first time and they receive the same payoff if they draw a blue ball the
second time they play the game. Although it violates expected and subjective expected utility, on average
participants draw both times from the box of which they know that it contains 50 blue and 50 red balls.
They strictly prefer drawing from the risky box of which they know that it represents a fair coin toss
over the box with the unkown distribution of balls.

Estimate
t-stat

Estimate
t-stat

Estimate
t-stat

lnDepth0

lnDepth3

Effective Spread

continued on next page . . .

Estimate
t-stat

Quoted Spread
0.0378a
(8.39)
0.0497a
(11.44)

0.0404a
(8.98)
0.0497a
(11.41)
-0.0323
(-1.39)

-0.0140
(-0.63)

-0.0257
(-1.08)

-0.0526c
(-2.09)

after news

before news

positive

0.0067
(1.71)
0.0947c
(2.59)

0.0811c
(2.28)

-0.0037
(-0.92)

(2.31)

0.0835c

after news

-0.0025
(-0.65)

-0.0117 b
(-2.95)

(2.22)

0.0794c

before news

negative

-0.0487 b
(-1.87)

0.0886a
(17.21)

0.0743a
(14.52)

(-3.10)

-0.0841 b

-0.0208
(-0.92)

0.0688a
(12.69)

0.0577a
(10.59)

-0.0463
(-1.82)

after news

neutral
before news

Panel A: Liquidity Estimations

-0.0183

0.0000

0.0026

-0.0269

before - after

positive

-0.0136

-0.0092

-0.0080

-0.0041

before - after

negative

-0.0278

0.0198

0.0167

-0.0379

before - after

neutral

Table 2.5: Trading Intensity and Liquidity Estimations Around News. Table 2.5 provides results for trading intensity and liquidity measures
around news in contrast to no-news periods over the years 2005 to 2008. The terms positive, negative, and neutral relate to the RNSE news
sentiment. The terms ‘before news’ and ‘after news’ describe thirty minute intervals before and thirty minute intervals after a news message
is disseminated over Thomson Reuters’ news wire systems. Overall GMM estimation results for liquidity measures are reported in Panel A
and for trading intensity in Panel B. All estimations are calculated with firm and year/quarter dummy variables. Robust t-statistics can be
found below estimates in parantheses. Quoted spreads and effective spreads are measured in basis points and ‘lnDepth0’ represents the natural
logarithm of the available volume at the best bid and ask in Canadian Dollars. ‘lnDepth3’ is the natural logarithm of available volume in
Canadian Dollars at the top three order book levels. Liquidity measures are aggregated to minute averages prior to estimation. The number of
shares traded per minute and the traded dollar volume per minute are transformed through the natural logarithm for the regressions. I omit
estimates for firm and time dummy variables. ‘a’ denotes significance at the 0.1% level, ‘b’ at the 1% level, and ‘c’ at the 5% level.

2.6 Results and Interpretation
41

Estimate
t-stat

Estimate
t-stat

Estimate
t-stat

#Trades per Min.

ln #Shares per Min.

ln Volume per Min.

. . . continued from Table 2.5

after news
1.6631a
(16.62)
0.1809a
(21.90)
0.1983a
(23.72)

before news
1.8201a
(18.09)
0.1890a
(22.84)
0.2053a
(24.49)

positive

0.1860a
(23.73)

0.2015a
(26.43)

2.8623a
(25.48)

before news

0.1885a
(24.13)

0.2956a
(26.80)

2.2139a
(20.90)

after news

negative

0.1614a
(17.77)

0.1555a
(17.39)

2.2551a
(18.56)

0.2787a
(27.64)

0.2769a
(28.15)

3.5241a
(24.99)

after news

neutral
before news

Panel B: Trading Intensity Estimations

0.0070

0.0081

0.1570

before - after

positive

-0.0025

-0.0042

0.6484

before - after

negative

-0.1172

-0.1214

-1.2690

before - after

neutral

42
2 High-Frequency Market Dynamics and Public Information

2.6 Results and Interpretation

43

2.6.2 Trading Intensity, Liquidity, and Volatility
To understand liquidity around news and to further understand the price dynamics around
news arrivals, it is important to bear in mind that liquidity, information, and trading intensity are inherently related. Table 2.5 provides regression results (Equation 2.12) on liquidity and trading activity. Results for liquidity are more mixed than for information.
However, compared with each other, all liquidity measures provide consistent results. I
find that before and after positive news liquidity increases. For negative news liquidity
generally falls, more so before than after news. Theory would suggest a consistent reduction in liquidity over all news types which is at odds with my empirical findings (cf. Kim
and Verrecchia, 1994). Consistent with existing literature (Berry and Howe, 1994), trading
intensity increases around all different types of news.
All coefficients in Table 2.5 represent the difference of the respective period to no-news
periods. Panel A shows that liquidity increases significantly around positive and around
neutral news. Liquidity increases if spreads tighten and depth increases. In this analysis
quoted spreads decrease 0.0526 bps before positive news and decrease 0.0257 bps after positive news, however not statistically significant after positive news. A decrease in quoted
spreads corresponds to an increase in liquidity. The liquidity enhancing effect is generally
stronger for neutral news than for positive. The increase in liquidity for neutral news is almost double the increase for positive news. The quoted spread increases, corresponding to
to a decrease in liquidity, before negative news by 0.0794 bps and 0.0835 bps after negative
news.
Results for Depth0, Depth3, and effective spreads are similar to those for quoted spreads.
All four measures combined paint a picture of increasing liquidity around positive and neutral news and decreasing liquidity around negative news. Not all measures are statistically
significant for all news types but when they are, values are highly coherent. For each news
type and each period before or after news at least two liquidity measures are statistically
significant and never contradictory. Comparing negative and positive news messages, the
former seem to have a stronger influence on spreads while the latter more strongly affect
available depth.
Panel B of Table 2.5 provides results for trading intensity. Trading intensity increases
around all types of news. It increases a bit more for negative than for positive news if
measured in the number of trades per minute and it increases stronger for positive news
if measured in Canadian dollar volume. However, trading intensity increases even more

44

2 High-Frequency Market Dynamics and Public Information

around neutral news compared to both positive and negative news. Substantial differential
interpretation by market participants can be observed which is in line with the neutrality
of such news messages; traders might not aggree on the meaning of neutral messages.
Table 2.7 provides in Panel A likelihood ratio (LR) tests and χ 2 statistics for all liquidity
and trading intensity estimations. All LR tests are highly significant which implies that all
models are better specified than the restricted models from the LR tests.
Since theory (cf. Kim and Verrecchia, 1991, 1994) predicts a reduction of liquidity
around information events my results for positive and neutral news may at first seem contradictory. However, the types of news that I analyze are different to the mostly studied scheduled macroceconomic announcements or earnings announcements. On average,
newswire messages surely have a lower impact than earnings or macroeconomic announcements and also their implications are on average much lower than those of major world
events. Trading intensity increases around news announcements which reflects changes in
expectations of individual investors who adjust to their new expectations through trade.
Liquidity suppliers in electronic limit order markets operate in a highly competitive environment (Biais et al., 1995). With higher trading intensity around positive and neutral
news announcements, liquidity suppliers compete for liquidity supply. They try to cater
the increase in liquidity demand. As explained above, traders potentially react differently
to positive and negative news messages. Reactions to positive and neutral news are weaker
than to negative news. Positive and neutral news might not be considered overly ambiguous by market participants such that competition for liquidity supply generally persists
around such news. Ranaldo (2006) also finds a slight increase in liquidity around news arrivals. Ambiguity aversion may help to explain the liquidity results around negative news.
If investors are expecting negative and ambiguous news, they will adjust their limit order
to include the ‘worst-case’ scenario that the negative news is precise. After negative news,
investors only slowly re-adjust their limit orders to the new information and no statistical
difference in liquidity to the no-news period can be found. Liquidity has also been found
to usually decrease around macroeconomic announcements (Fleming and Remolona, 1999;
Green, 2004). A single negative news has on average still much less impact and importance
in comparison to e.g. macroeconomic news. However, in terms of a trader’s perception
of the strenght of impact, negative news might be potentially closer to macroeconomic
news or earnings announcements than positive news such that they are considered more
important and reactions are stronger (asymmetric reaction). As a result, competition for
liquidity supply might not increase around negative news messages but even slightly falls.

2.6 Results and Interpretation

45

Additionally, I find that realized volatility is slightly higher around arrivals than during
no news periods consistent over all types of news: positive, negative, and neutral. Table 2.6
provides the exact coefficients for realized volatility around news arrivals including robust
t-statistics. The LR test χ 2 statistic for the realized volatility estimation is 34 which is
highly significant at the 0.01% level (Table 2.7). The results for realized volatility pre- and
post-news arrival are also consistent with the MRR information results. Around news,
adverse selection costs are higher than during no-news periods for all three different news
sentiments (cf. Table 2.4, Panel B) which indicates higher private information flow around
news. French and Roll (1986) find that a major determinant of return volatility is trading of
informed market participants, i.e. private information flow revealed to the market through
trades. I find this pattern in the data of this chapter with higher realized volatility around
news.

Realized Vola.

Estimate
t-stat

after news
0.0014a
(5.99)

before news
0.0016a
(6.21)

positive

(9.50)

0.0031a

before news
(4.78)

0.0015a

after news

negative

(5.04)

0.0018a

(8.65)

0.0032a

after news

neutral
before news

Realized Volatility Estimations

0.0003

before - after

positive

0.0017

before - after

negative

-0.0014

before - after

neutral

Table 2.6: Realized Volatility Estimations Around News. Table 2.6 provides estimates for realized volatility measures around news in
contrast to no-news periods over the years 2005 to 2008. The terms positive, negative, and neutral relate to the RNSE news sentiment. The
terms ‘before news’ and ‘after news’ describe thirty minute intervals before and thirty minute intervals after a news message is disseminated
over Thomson Reuters’ news wire systems. All estimations are calculated with firm and year/quarter dummy variables. Robust t-statistics
can be found below estimates in parantheses. Realized Volatility is based on one-minute returns. I omit estimates for firm and time dummy
variables. ‘a’ denotes significance at the 0.1% level, ‘b’ at the 1% level, and ‘c’ at the 5% level.

46
2 High-Frequency Market Dynamics and Public Information

2.6 Results and Interpretation

47

Table 2.7: Likelihood Ratio Tests. Table 2.7 provides likelihood ratio (LR) test statistics for the
estimations for liquidity, trading intensity, realized volatility, and the MRR model. The likelihood
test results provide statistics for the restricted model with all pre- and post-news intervals captured
by one coefficent. Panel A provides χ 2 statistics for the liquidity, trading intensity, and realized
volatility models and p-values. Panel B provides likelihood ratio test statistics for the MRR model.
The likelihood test results provide statistics for the restricted MRR model with all pre-, post- and
no-news periods captured by one coefficent.
Panel A: LR Tests
χ 2 -stat

p-value

Liquidity
Quoted Spread
lnDepth0
lnDepth3
Effective Spread

28
290
310
18

0.0002
< .0001
< .0001
0.0029

Trading Intensity
#Trades per Min.
ln #Shares per Min.
ln Volume per Min.

161
93
84

< .0001
< .0001
< .0001

Realized Volatility

34

< .0001

Panel B: LR Tests for MRR Estimations

Adverse Selection θ
Order Processing φ
Autocorrelation ρ

Mean
χ 2 -stat

Median
χ 2 -stat

# of significant
out of 33 (0.1% level)

240
260
281,331

194
189
111,977

32
32
33

2.6.3 Robustness
Since the main sample period includes the financial crisis period in 2008, I also perform an
analysis on a control data set. The control sample comprises of the same firms as the main
sample. Only for the MRR estimations, the number of firms reduces from 33 to 26 since I
apply the same rules for a minimum number of news as in the main sample.22 A minimum
number of news messages is required to receive stable MRR estimation results. My control data observation period comprises the years 2003 to 2006, four years comparable to the
main observation period. The year 2003 is the first year with RNSE archive data available.
The years 2003 to 2006 encompass a financially stable period. Figures 2.4, 2.5, and 2.6 de22

The following firms are removed from the original sample for the robustness check MRR sample as a
result of an insufficient number of news messages: Agnico-Eagle Mines Ltd. (AEM.TO), Agrium Inc.
(AGU.TO), Magna International Inc. (MGa.TO), Potash Corporation of Saskatchewan Inc. (POT.TO),
and Rogers Communications Inc. (RCIb.TO).

2 High-Frequency Market Dynamics and Public Information

48

250

200

150

100

50

0

2003

2004

2005

2006

Figure 2.4: Novel Intraday News Per Year and Month on the TSX 2003 to 2006. The figure
shows the number of novel intraday news messages per year and month for the 2003 to 2006 sample.

pict the number of news per month of a year, the number of news per weekday, and the
number of news per time of day. The number of news over the robustness period is not
dramatically different from the main sample period with 5,590 news messages and 6,625
news messages respectively. The estimations for information, liquidity, and trading intensity are performed exactly like the estimations for the 2005 to 2008 period. Estimations
also include dummies for day of the week and time of day effects. Tables 2.8 and 2.9 show
the estimation results.

2.6 Results and Interpretation

49

1,400
1,200
1,000
800

neutral
negative
positive

600
400
200
0

Mon

Tue

Wed

Thu

Fri

Figure 2.5: Novel Intraday News Per Weekday on the TSX 2003 to 2006. The figure shows the
number of novel intraday news messages per day of the week for the 2003 to 2006 sample.

2 High-Frequency Market Dynamics and Public Information

50

800
700
600
500
neutral
negative
positive

400
300
200
100
0

1st 2nd 3rd 4th 5th 6th 7th 8th 9th 10th 11th 12th 13th

Figure 2.6: Novel Intraday News Per Time of Day (half-hours) on the TSX 2003 to 2006. The
figure shows the number of novel intraday news messages for time of day half-hour intervals for
the 2003 to 2006 sample.

Median Est.
Median t-stat
Median Est.
Median t-stat

Order Proc.
φ

Autocorr.
ρ

continued on next page . . .

Median Est.
Median t-stat

Adv. Selection
θ

0.3286
(367.71)

1.5190
(177.60)

1.2855
(200.37)

no news (nn)

0.4212
(34.76)

1.2682
(8.76)

1.9038
(13.18)

before news

0.4579
(32.49)

1.2844
(8.55)

1.9451
(14.73)

after news

positive

0.4972
(39.11)

1.1947
(5.98)

2.4117
(11.93)

before news

0.3463
(43.92)

0.7202
(2.99)

2.3536
(12.72)

after news

negative

Panel A: By-Company Information Estimations - MRR Model

0.4124
(25.58)

1.2626
(8.01)

1.8432
(13.55)

before news

0.5313
(29.84)

0.7705
(6.36)

2.1576
(12.77)

after news

neutral

Table 2.8: Information Estimations Around News – Control Period. Table 2.8 provides the results of the MRR model for no-news and
news periods. Results comprise of the adverse selection components θ, order processing costs φ, and trade autocorrelations ρ. The terms
positive, negative, and neutral relate to the RNSE news sentiment. ‘Before news’ and ‘after news’ describe thirty minute intervals before and
thirty minute intervals after a news message is disseminated over Thomson Reuters’ news wire systems. The MMR model is estimated on a
per company basis for the years 2003 to 2006. By-company estimation results in Panel A consist of the medians and means of GMM estimation
results for each single company in the sample. Robust median t-statistics can be found below estimates in parantheses. Panel B provides
differences between different intervals and no-news periods and between pre- and post-news periods. The medians of the differences ∆Est are
compared with Wilcoxcon Signed Rank tests. ‘a’ denotes significance at the 0.1% level, ‘b’ at the 1% level, and ‘c’ at the 5% level.

2.6 Results and Interpretation
51

-0.2965 b
0.0071
0.0810a
< .0001

Median ∆Est
p-value
(Wilcoxon T.)

Median ∆Est
p-value
(Wilcoxon T.)

Median ∆Est
p-value
(Wilcoxon T.)

Adv. Selection
θ

Order Proc.
φ

Autocorr.
ρ

0.1092a
< .0001

-0.1338c
0.0112

< .0001

0.2677a

< .0001

after - nn

0.2578a

positive

0.1680a
< .0001

-0.3429c
0.0156

< .0001

0.7150a

before - nn

0.2445a
< .0001

-0.5504 b
0.0011

< .0001

0.7269a

after - nn

negative

0.0649a
0.0004

-0.3396 b
0.049

< .0001

0.3500a

before - nn

0.2006a
< .0001

-0.5054a
< .0001

< .0001

0.6248a

after - nn

neutral

Panel B: By-Company Information Estimation Differences - MRR Model

before - nn

. . . continued from Table 2.8

-0.0027
0.8726

0.0276
0.9508

0.0840
0.3641

before - after

positive

-0.0064
0.5116

0.2484
0.1791

-0.0490
0.3776

before - after

negative

-0.0593
0.0816

0.2542
0.2056

-0.2353
0.1403

before - after

neutral

52
2 High-Frequency Market Dynamics and Public Information

Estimate
t-stat

Estimate
t-stat

Estimate
t-stat

lnDepth0

lnDepth3

Effective Spread

continued on next page . . .

Estimate
t-stat

Quoted Spread

after news
-0.2908a
(-9.44)
0.0077
(1.65)
0.0281a
(7.10)
-0.2487a
(-8.88)

before news
-0.2939a
(-9.25)
0.0199a
(4.29)
0.0287a
(7.14)
-0.2397a
(-8.30)

positive

-0.3769a
(-12.64)

-0.0100c
(-2.36)

-0.0187a
(-3.87)

(-14.61)

-0.4712a

before news

-0.3175a
(-10.66)

-0.0076
(-1.87)

-0.0215a
(-4.49)

(-12.08)

-0.3810a

after news

negative

-0.1129c
(-2.21)

0.0094
(1.79)

-0.0013
(-0.22)

(-3.02)

-0.1609c

-0.2128a
(-5.32)

0.0128c
(2.43)

0.0084
(1.40)

-0.1462
(-1.85)

after news

neutral
before news

Panel A: Liquidity Estimations

0.0090

0.0007

0.0122

-0.0031

before - after

positive

-0.0594

-0.0023

0.0028

-0.0903

before - after

negative

0.1000

-0.0035

-0.0097

-0.0147

before - after

neutral

Table 2.9: Trading Intensity and Liquidity Estimations Around News – Control Period. Table 2.9 provides results for trading intensity and
liquidity measures around news in contrast to no-news periods over the years 2003 to 2006. The terms positive, negative, and neutral relate to
the RNSE news sentiment. The terms ‘before news’ and ‘after news’ describe thirty minute intervals before and thirty minute intervals after a
news message is disseminated over Thomson Reuters’ news wire systems. Overall GMM estimation results for liquidity measures are reported
in Panel A and for trading intensity in Panel B. All estimations are calculated with firm and year/quarter dummy variables. Robust t-statistics
can be found below estimates in parantheses. Quoted spreads and effective spreads are measured in basis points and ‘lnDepth0’ represents the
natural logarithm of the available volume at the best bid and ask in Canadian Dollars. ‘lnDepth3’ is the natural logarithm of available volume
in Canadian Dollars at the top three order book levels. Liquidity measures are aggregated to minute averages prior to estimation. The number
of shares traded per minute and the traded dollar volume per minute are transformed through the natural logarithm for the regressions. I omit
estimates for firm and time dummy variables. ‘a’ denotes significance at the 0.1% level, ‘b’ at the 1% level, and ‘c’ at the 5% level.

2.6 Results and Interpretation
53

Estimate
t-stat

Estimate
t-stat

Estimate
t-stat

#Trades per Min.

ln #Shares per Min.

ln Volume per Min.

. . . continued from Table 2.9

after news
0.9213a
(12.67)
0.1892a
(18.94)
0.2107a
(20.25)

before news
0.9364a
(14.17)
0.2074a
(20.66)
0.2242a
(21.46)

positive

0.2776a
(23.83)

0.2552a
(23.10)

1.5297a
(18.30)

before news

0.2623a
(23.21)

0.2405a
(22.45)

1.4126a
(18.35)

after news

negative

0.1460a
(10.91)

0.1373a
(10.77)

0.8925a
(9.62)

0.2908a
(21.17)

0.2718a
(20.64)

1.3618a
(13.95)

after news

neutral
before news

Panel B: Trading Intensity Estimations

0.0134

0.0182

0.0151

before - after

positive

0.0153

0.0147

0.1170

before - after

negative

-0.1449

-0.1346

-0.4693

before - after

neutral

54
2 High-Frequency Market Dynamics and Public Information

2.6 Results and Interpretation

55

Comparable to the main observation period, adverse selection costs increase significantly around news arrivals. The increase around negative news is much stronger than
around positive news. The only difference is a strong increase after the arrival of neutral
news. An increase in adverse selection costs after neutral news is generally consistent with
the main observation period but the magnitude of the increase is much higher in comparison to other news types and periods. However, it is also not entirely clear what drives the
results for adverse selection after neutral news during the main observation period.
Liquidity results are more mixed than in the main observation period. While results
are very consistent for positive and neutral news with a liquidity enhancing effect, results
are more ambiguous around negative news messages. One can see a clear indication of
a reduction of liquidity around negative news in the main observation period from 2005
to 2008. In the control period, one cannot say whether liquidity increases or decreases
around negative news messages. Spreads decrease while depth also decreases. What can be
said about negative news, again consistent with the main observation period, is that I do
not find a clear increase in liquidity in contrast to positive and neutral news. A possible
explanation for the differences might be that liquidity suppliers react more sensitive to
negative news during the financial crisis period. Trading intensity results in the control
period are consistent with the main observation period. Trading intensity increases around
all types of news. Table 2.10 provides LR tests for the control period which show that the
models are better specified than the restricted models.

2.6.4 Returns and Profitability
The previous sections provide evidence that there is a significant difference between positive, negative, and neutral news in terms of price discovery and liquidity. In this section, I
provide a simple profitability analysis based on groups of news with different sentiments.
All returns z g in this section are calculated against the market index, the TSX 60. I calculate
the 10 minute returns pre- and post-news arrivals and the return from 10 minutes before a
news message is released to 10 minutes after a news message arrives. The same analysis is
performed for thirty minute intervals. From a profitability perspective post-news returns
are most interesting. The independent variable in the regression (see subsection 2.5.3) is
the sentiment of a news message multiplied with the relevance of a news message. I hypothesize that the impact on returns is stronger if a news message is more relevant for an
instrument. For each time interval (10 minutes or 30 minutes), four different groups of

56

2 High-Frequency Market Dynamics and Public Information

Table 2.10: Likelihood Ratio Tests – Control Period. Table 2.10 provides likelihood ratio (LR)
test statistics for the estimations for liquidity, trading intensity, realized volatility, and the MRR
model. The likelihood test results provide statistics for the restricted model with all pre- and postnews intervals captured by one coefficent. Panel A provides χ 2 statistics for the liquidity, trading
intensity, and realized volatility models and p-values. Panel B provides likelihood ratio test statistics
for the MRR model. The likelihood test results provide statistics for the restricted MRR model with
all pre, post and no-news periods captured by one coefficent.
Panel A: LR Tests
χ 2 -stat

p-value

Liquidity
Quoted Spread
lnDepth0
lnDepth3
Effective Spread

38
59
82
28

< .0001
< .0001
< .0001
< .0001

Trading Intensity
#Trades per Min.
ln #Shares per Min.
ln Volume per Min.

64
82
87

< .0001
< .0001
< .0001

Panel B: LR Tests for MRR Estimations

Adverse Selection θ
Order Processing φ
Autocorrelation ρ

Mean
χ 2 -stat

Median
χ 2 -stat

# of significant
out of 26 (0.1% level)

266
328
93,954

128
166
28,585

26
24
26

messages are analyzed to isolate the group with the highest effect on returns. I examine
all news, all news without positive news, all news without negative news, and all news
without neutral news. Table 2.11 reports all regression results on returns. There are no
significant coefficients for the 10 minute intervals around news. The picture changes for 30
minute intervals. Results show that excess returns are driven by negative messages. I find
significant coefficients at the 10% level in all groups except the one without negative news.
Significant returns only exist for time intervals that include the post-news period. These
results provide some evidence that returns are indeed driven by news messages, i.e. public
information. If accounted for transaction costs, returns based on the simple strategie to
buy and sell solely based on the product of sentiment and relevance are not high enough to
sustain a profitable business. However, highly sophisticated automated trading strategies
based on news might have the potential to generate sustainable positive returns which are
profitable from a business perspective.

2.7 Conclusion

57

2.7 Conclusion
In this chapter, I analyze the impact of Thomson Reuters newswire messages on intraday
price discovery, liquidity, and trading intensity at the Toronto Stock Exchange. In contrast
to existing literature, I am able to cluster news based on message content. News data are
split into groups of news messages with positive, negative, and neutral sentiment which
gives me the opportunity to study asymmetric reactions to news messages. News messages
are not sorted based on ex-post return measures but on ex-ante message content based
measures. The adverse selection component of the spread is estimated with an extension
of the Madhavan et al. (1997) model.
Results provide evidence of asymmetric reactions to news. In general, I find higher
adverse selection costs around news messages which can be explained through information gathering prior to news arrivals and differential information processing capabilities of
market participants after news arrivals. On the the sentiment level, negative news messages
induce significantly higher adverse selection costs than positive news messages. Liquidity
increases around positive and neutral messages whereas it decreases around negative messages. Trading intensity increases around all types of news messages. A possible explanation for the difference between news messages with different sentiment could be ambiguity
aversion and asymmetric reaction to news. Ambiguity averse traders react asymmetrically
to ambiguous information such as news messages. If the market is composed of a proportion of ambiguity averse traders, this provides a possible explanation for my results.
The main contribution of this chapter is that I show that traders react asymmetrically to
intraday news arrivals. I find that newswire messages as one form of public information
generally have a significant impact on intraday trading in an electronic limit order market.
The next chapter studies the impact of news in a different institutional setting with high
market fragmentation and public information’s impact on fragmentation characteristics.
In contrast to this chapter, information, liquidity, and trading activity measures are aggregated to daily averages to facilitate the comparison of different markets.

2 High-Frequency Market Dynamics and Public Information

58

Table 2.11: Profitability Analysis. Table 2.11 provides basic profitability analyses. Excess returns
are calculated for each stock with the TSX/S&P 60 index returns for 10 and 30 minute intervals
around news arrivals. I regress returns on sentiment multiplied by relevance and include company
dummy variables. Regressions are performed on all news and on groups of news each missing
either positive, negative, or neutral news messages. Robust t-statistics are provided in parentheses.
‘a’ denotes significance at the 1% level, ‘b’ at the 5% level, and ‘c’ at the 10% level.
Excess Returns (z g ) All News

Pre News
Post News
Pre and Post

10 min. around news

30 min. around news

Estimate

t-stat

Estimate

t-stat

-0.000030
-0.001200
-0.001140

(-0.26)
(-0.82)
(-0.83)

0.001235
0.002701c
0.000753c

(1.10)
(1.88)
(1.69)

Excess Returns (z g ) No Positive News

Pre News
Post News
Pre and Post

10 min. around news

30 min. around news

Estimate

t-stat

Estimate

t-stat

0.000154
-0.002170
-0.002030

(0.70)
(0.79)
(-0.73)

0.005476
0.005933c
0.001104c

(1.46)
(1.75)
(2.40)

Excess Returns (z g ) No Negative News

Pre News
Post News
Pre and Post

10 min. around news

30 min. around news

Estimate

t-stat

Estimate

t-stat

-0.000050
-0.000080
-0.000100

(-0.31)
(-0.55)
(-0.42)

-0.001960
-0.001280
0.000239

(-0.92)
(-0.97)
(0.66)

Excess Returns (z g ) No Neutral News

Pre News
Post News
Pre and Post

10 min. around news

30 min. around news

Estimate

t-stat

Estimate

t-stat

-0.000020
-0.001080
-0.001100

(-0.20)
(-0.81)
(-0.82)

0.001052
0.002410c
0.000737c

(0.98)
(1.77)
(1.66)

Chapter 3
Fragmented Markets and Public
Information
3.1 Introduction
Both, technology and regulation, have radically changed trading in equities. The dramatic
advancement of information and communication technology has enabled providers of
trading venues to operate their markets entirely electronically in computing centers without any floor interaction. The competitive barriers of entering the market as a trading
venue operator have significantly decreased as a result of technology. Additionally, regulation in Europe1 and the United States2 has allowed for new regulated trading venues in
addition to incumbent exchanges. Such alternative trading venues are called multilateral
trading facilities (MTF) in Europe and electronic communication networks (ECN) in the
United States. Incumbent exchanges have lost signficant market shares to such alternative
trading venues. As a consequence, specifically in Europe, trading in blue chip stocks is no
longer concentrated on one national exchange.
In such a dynamic trading landscape, new information has multiple opportunities to
translate into prices. In this chapter, I analyze the impact of the tone found in firm specific
public information proxied through Thomson Reuters newswire messages on trading in
fragmented markets in FTSE 100 constituents listed on the London Stock Exchange (LSE).
Within the scope of the analysis, two fundamental research questions arise. First, what is
1

In Europe, the Markets in Financial Instruments Directive (MiFID) allows for additional regulated trading
venues.
2
In the United States, Regulation NMS determines how orders are handeled between regulated exchanges
and alternative trading venues with the goal to foster competition.

3 Fragmented Markets and Public Information

60

the impact of positive or negative firm specific public information on trading, specifically
liquidity and information, in individual electronic securities markets in a fragmented market environment. Second, how does positive or negative firm specific public information
influence characteristics of market fragementation and traders’ preferences for different
markets. In contrast to much of the existing public information literature, I use an ex-ante
measure for the tone of public information. My data set of newswire messages includes,
since it is the same as in Chapter 2, a computed sentiment measure which is either positive,
negative, or neutral for a single news item with respect to a specific firm. In contrast to
Chapter 2, the final regression framework analyzes measures on a daily basis to enable a
comparison of different trading venues. However, daily averages are still compiled from
intraday trading data to capture market microstructure effects.
The main results of this chapter are that I find on negative public information days
lower liquidity, an increase in trading activity especially in mid-sized trades on the LSE,
and a small growth in private information. I also observe a shift in private information
processing to the LSE as a result of negative public information. On days with positive
public information, no significant change in liquidity is discovered, again a strong rise
in trading activity, and overall less private information impounded into markets but a
significant shift of the remaining private information from Chi-X to the LSE. One key
finding is that negative and positive public information have an asymmetric impact on
trading. Also, informed trading resorts to the LSE during times of high levels of public
information consistent with Chowdhry and Nanda (1991). This result is consistent with
literature which shows that market participants trade off factors such as execution speed
for liquidity and flexibility under uncertain market conditions (Goldstein and Kavajecz,
2004).
The remainder of this chapter is structured as follows. Section 3.2 introduces related
work. Section 3.3 gives an overview of the institutional structures of the LSE and Chi-X.
Section 3.4 provides a description of the newswire data set, trading data, and the sample
while Section 3.5 presents market measures used in this chapter. Section 3.6 introduces the
regression framework and provides results and Section 3.7 finally concludes this chapter.

3.2 Related Work
Existing public information literature considers different types of public information
ranging from media content to scheduled earnings announcements. Thomson Reuters

3.2 Related Work

61

newswire messages are somewhat in-between those extrems. In addition, in the securities
trading industry, newswire messages represent a large portion of the real-time information
traders receive.
Many papers that investigate the effect of public information or ambiguous linguistic
news content on financial markets are already presented in Chapter 2 Section 2.2. One
study by Ryan and Taffler (2007) specifically analyzes trading and public information at
the LSE. The authors find that firm specific news releases drive trading activity in the
British market especially in FTSE 100 trading. The major news source that drives trading
volume in their study is analyst activity measured through analyst reports in news data.
They argue that analyst activity also represents “sell-side analysts possessing superior information processing skills and/or having access to ‘private’ information”. In general, an
increase in trading volume reflects differential interpretation of information by investors
(Kim and Verrecchia, 1991). Traders disagree and as a result shift to their new expectation level, which is based on the additional information, through trade. The relation of
trading volume and public information is a well described phenomenon, also in papers presented in Chapter 2. Morse (1981) provides one very early study based on daily data. He
shows that earnings announcements significantly increase daily trading volume. Mitchell
and Mulherin (1994) examine the number of daily news announcements of Dow Jones &
Company. They find a direct, however small, relation between trading activity and the
number of Dow Jones messages. Empirical evidence for the price discovery process and
liquidity is not as clear as it is for trading volume and existing studies report sometimes
conflicting results (cf. Chapter 2 Section 2.2). In contrast to existing studies that focus on
returns, this chapter focuses on the influence of firm specific public information on explicit
market microstructure characteristics, i.e. liquidity, trading activity, and information.
From a theoretical perspective the, in Chapter 2 presented, Kim and Verrecchia (1991,
1994) models provide predictions about adverse selection costs, trading activity, and liquidity for a single market. Pre-announcement information gatherers have superior private
information which increases the adverse selection component of the spread while the same
effect is observed post-announcement as a result of different information processing capabilities of market participants. Additionally, the models predict an increase of trading
volume and a decrease in liquidity around public information announcements. However,
the Kim and Verrecchia (1991, 1994) models do not provide predictions about what should
happen between markets if an instrument is traded on several trading venues. For my analysis, existing empirical evidence and financial theory suggest for an individual market that

62

3 Fragmented Markets and Public Information

I find an increase in trading activity on days with a high level of public information. Empirical evidence for liquidity and adverse selection is mixed while financial theory suggests
an increase in adverse selection costs and a reduction of liquidity.
On 1 November 2007, MiFID, the Markets in Financial Instruments Directive3 , passed
by the European Union, came into effect. The stated “objective of MiFID is to foster a fair,
competitive, transparent, efficient, and integrated European financial market” (Degryse,
2009). Market fragmentation in Europe is a relatively new phenomenon. Prior to MiFID,
a single incumbent exchange existed in most European countries with little competition
to fear. EU countries had the possibility to employ a concentration rule which required
that all orders had to be executed on a regulated market.
MiFID has enabled alternative trading venues, MFTs, to compete against regulated exchanges. Specifically, an MTF is defined as “a multilateral system, operated by an investment firm or a market operator, which brings together multiple third-party buying and
selling interests in financial instruments – in the system and in accordance with nondiscretionary rules – in a way that results in a contract...”4 . Under MiFID, MTFs are
regulated by the national regulatory authorities. MTFs challenge incumbent exchanges
with fast trading platforms, innovative order types, and innovative fee systems. In contrast to US regulation5 , markets are not formally linked and best execution obligations
are not primarily focused on the best price principle but include multiple dimensions. Investment firms must “take all reasonable steps to obtain, when executing orders, the best
possible result for their clients taking into account price, costs, speed, likelihood of execution and settlement, size, nature or any other consideration relevant to the execution
of the order”6 . Within MiFID, best execution obligations have to be met by financial intermediaries not exchanges. Often, incumbent exchanges stress that investors enjoy less
protection when trading on an MTF than they do on incumbent regulated markets. But
under MiFID, both, MTFs and regulated markets, are regulated by national regulatory authorities and need adhere to similar rules such as “transparent and non-discretionary rules
3

Directive 2004/39/EC.
Article 4(1) 2004/39/EC.
5
Rule 611 of Regulation NMS, the order protection rule, defines that national US markets are protected
against trade-throughs. A trade-through occurs when a trade is executed on one market despite the fact
that another national market offers a better quote. The SEC’s rationale behind the order protection rule
is that a protection against trade-throughs would incentivize market participants to post limit orders and
thus supply liquidity. The order protection rule only protects the top of the book not the depth of the
book.
6
Article 21(1) 2004/39/EC.
4

3.2 Related Work

63

and procedures for fair and orderly trading”7 . As fast and technologically reliable platforms
with competitive pricing, MTFs have been especially successful in attracting order flow for
trading in FTSE 100 stocks. On some days in the the first quarter of 2010, the LSE had a
market share of less than 50% in its blue chip segment.
To my knowledge there is currently no study that directly investigates the effect of public information specifically in fragmented markets and on characteristics of market fragmentation. With the recent developments in regulation8 , I am interested in how firm
specific public information influences trading in fragmented markets. With high market
shares, alternative markets are important to be included in empirical analyses.
A recent study by Riordan et al. (2010a) examines market fragmentation in FTSE 100
stocks with an analysis of the LSE and the three largest MTFs (Chi-X, Turqoise, and
BATS). They find that the major markets are the LSE and Chi-X with Turquoise and
BATS having little influence on price discovery. Chi-X leads in quote based price discovery whereas the LSE leads in trade based price discovery. Both, the LSE and Chi-X, are
highly liquid and contribute significantly to total price discovery in FTSE 100 stocks.
Other evidence on market fragmentation is mixed. Mendelson (1987) and Bennett and
Wei (2006) find lower liquidity and less efficient markets as a result of the fragmentation
of order flow. Mendelson (1987) provides a theoretical model to assess the influence of
market fragmentation on price discovery in different market microstructure settings. The
author also provides a concise definition of market fragmentation. “We say that the market
mechanism is consolidated if all the order data are available when this transformation [of
orders to transactions] takes place, e.g., when all orders are channeled to a central trading
post. We say that the market is fragmented when orders are decomposed into a number of
disjoint subsets, and the transformation is applied to each subset separately, e.g., when an
asset is traded in a number of secluded locations” (Mendelson, 1987). He finds that fragmentation can reduce trading volume and increase volatility. However, he concludes that
there is no per se optimal solution. “The diversity of exchange mechanisms that prevail
around the world as well as across assets reflects the dependence of the appropriate market
design on specific circumstances and on factors that are probably not captured by the stylized facts of the market microstructure literature” (Mendelson, 1987). His study calls for a
careful analysis of the effects of fragmentation in a specific market setting before jumping
7
8

Article 14(1) 2004/39/EC.
MiFID is currently under review and the European Commission has opened consultation on the MiFID
review. After the consultation period, the European Commission will propose a MiFID amandement in
spring 2011 (http://ec.europa.eu/internal_market/consultations/2010/mifid_en.htm).

3 Fragmented Markets and Public Information

64

to conclusions, for instance with policy decisions. Bennett and Wei (2006) observe liquidity and price discovery for stocks that switch from Nasdaq to the NYSE in the years 2002
and 2003. In those years, order flow of Nasdaq listed stocks was substantially more fragmented than order flow of NYSE listed stocks. The authors attribute a liquidity and price
discovery enhancing effect to the increase in order flow consolidation after the switch to
the NYSE.
Boehmer and Boehmer (2003), Barclay et al. (2003), and Goldstein et al. (2008) ascertain that competition for order flow has positive effects on price discovery and liquidity.
Boehmer and Boehmer (2003) study a natural experiment in 2001 and 2002 when AMEX
listed ETFs start trading on the NYSE which results in a higher fragmentation of the order flow. Within the first month of trading, the NYSE has gained a market share of 10%.
The ETFs previously traded on AMEX, Nasdaq, and Island ECN, with AMEX having a
very similar market structure as the NYSE. Both measured in spreads and depth, market
specific liquidity as well as consolidated liquidity strongly increase after the ETFs start
trading on the NYSE. Barclay et al. (2003) investigate the interaction of price discovery
and liquidity of stocks which are listed on Nasdaq and traded on both Nasdaq and Electronic Communication Networks (ECN). Their data is from the year 2000 when ECNs
already executed a major share of order flow of Nasdaq listed stocks. They find that a
higher fraction of informed trading is executed on ECNs than on Nasdaq. Those results
should counter the often raised concern that ECNs only cream skim order flow when in
fact they contribute substantially to price discovery. Goldstein et al. (2008) study Nasdaq
listed stocks in a more recent period than Barclay et al. (2003) with data from 2003. In
their paper “quote competitiveness is found to increase the probability of executions on
all four venues[, Nasdaq, Archipelago, Instinet, and Island ECN,]” (Goldstein et al., 2008)
however they conclude that extreme competition among trading venues could be harmful
in the long run especially for small cap stocks.
Foucault and Menkveld (2008) combine a theoretical model and empirical analysis in
one paper. Their model predicts that market fragmentation and competition result in
higher consolidated liquidity and that the liquidity supply of the new trading venue increases with smart order routing. They approximate smart order routing in their empirical
analysis with the fraction of trades that do not violate price priority. In the empirical part,
they confirm their model with an analysis of the Dutch stock market after the 2004 introduction of EuroSETS, an alternative trading venue which competes against the Euronext
system. First, they find more liquidity in the consolidated order book. They also find

3.3 Institutional Details

65

that trade throughs, the violation of price priority across order books, discourage limit orders. With a higher fraction of traders using smart order routing, trade throughs occur less
and EuroSETS provides more liquidity. As a policy implication, Foucault and Menkveld
(2008) conclude that is is important to have some protection against trade throughs.
In finance theory, models exists that predict that investors have incentives to concentrate
order flow on one market. Pagano (1989) develops a multimarket model in which identical
execution costs lead to a concentration on one market. However, investors’ preferences
might differ in real market environments which could lead to a sustainable multi-market
solution. One example could be large liquidity traders that trade on special trading venues
to circumvent larger adverse price movements. In this chapter, I am specifically interested
in how information is processed and how characteristics of fragmented markets vary with
high levels of public information. From a theoretical point of view, Chowdhry and Nanda
(1991) model that informed trading gravitates to the most liquid market. Informed traders
have the opportunity to use their private information in multiple markets, however, they
are attracted by liquidity. If public information processors have higher levels of private
information as a result of public information, one would expect that trading volume and
price discovery shifts to the more liquid market. Since most trading in FTSE 100 stocks
is still executed on the LSE and the LSE’s daily trading volume is much larger than that
of Chi-X, I anticipate that volume and price discovery shifts to the LSE in times of high
levels of public information.

3.3 Institutional Details
In this analysis, I study two markets that offer trading in FTSE 100 stocks: the LSE and
Chi-X. The LSE is the incumbent exchange on which FTSE 100 constituents are listed.
It is one of the world’s largest equity exchanges with an annual value of share trading
of 2,796,077 mGBP9 whereas Chi-X is a multilateral trading facility (MTF) which has
emerged only recently but has increased its market share steadily. Both, the LSE and
Chi-X, are regulated through the Financial Services Authority (FSA) which is the British
regulator of the financial services industry, the LSE as a regulated exchange under MiFID
and Chi-X as an MTF. Chi-X started trading about six month ahead of MiFID at the end
of March 2007. The full list of FTSE 100 constituents became available on Chi-X on 13
9

World Federation of Exchanges,
http://www.world-exchanges.org/statistics/annual/2009/equity-markets/total-value-share-trading/.

3 Fragmented Markets and Public Information

66

July 2007. Currently, Chi-X is owned by Instinet, a subsidiary of Nomura Holding, and a
number of international investment banks. Its market share in UK equity trading has increased from 9% in March 2008 to 15% when it celebrated its second anniversary in March
200910 , and finally to 26% in in the first quarter of 201011 . Currently, Chi-X is the largest
among all existing MTFs.12 Chi-X not only offers trading in British FTSE 100 constituents
but also trades, for instance, stocks listed in the French CAC 40, the German DAX 30, and
the Dutch AEX 25.
Chi-X and the LSE compete predominantly on technology and trading costs which
translates into different fee structures, network latencies, and IT system service levels.
FTSE 100 constituents are traded on the SETS13 system at the LSE which provides a combination of an electronic limit order market with liquidity provision through market makers.
Market makers operate within the electronic public limit order book without proprietary
information. Liquidity for non-crossing orders is solely provided by limit orders displayed
in the order book. Orders are executed with price-time priority. Iceberg orders that display only a portion of their total size sacrifice time priority on the non-displayed portion
of the order such that the priority rule could more precisely be called price-visibility-time
priority. The LSE introduced hidden liquidity only on 14 December 2009 to match MTFs.
Such orders have to meet the Large-In-Scale considerations of MiFID.14 These types of orders add liquidity to an order book and are primarily used by informed investors to avoid
adverse selection costs.
Chi-X also operates an entirely electronic limit order book with a combination of visible
and hidden liquidity based on price-time priority. Comparable to the LSE, hidden orders
sacrifice their time priority and have to meet MiFID’s Large-In-Scale requirements. In
addition to limit orders, market orders, iceberg orders, and hidden orders, Chi-X offers
pegged orders. The trading price for such orders is determined based on a reference price,
for instance the European Best Bid and Offer (EBBO). Orders on Chi-X are subject to a
price check to ensure investors that orders are not executed far away from prices above
or below the European Best Bid and Offer. Technically, Chi-X has, on average, ten times
lower latencies with 0.4 ms than the LSE.
10

http://www.chi-x.com/chi-x-press-releases/Chi-X-Europe-Second-Year-Anniversary.pdf.
http://www.chi-x.com/chi-x-press-releases/chi-x-europe-q1-2010-trading-stats-final.pdf.
12
Other MTFs are, for instance, Turquoise and BATS.
13
Stock Exchange Electronic Trading Service, http://www.londonstockexchange.com/traders-andbrokers/products-services/trading-services/sets/sets.htm.
14
MiFID, Directive 2004/39/EC Article 22(2).
11

3.4 Data and Sample Selection

67

During most of the observation period, the LSE and Chi-X both feature a maker-taker
pricing scheme. At the LSE an investor is charged between 0.45 basis points (bps) to 0.75
bps for an aggressive order, an incoming order which hits an existing order in the order
book. Executed passive orders receive a rebate of up to 0.40 bps. The maker-taker fees and
rebates depend on monthly executed order volume. The highest rebate is received above a
monthly trading volume of 25 bnGBP, the minimum fee of 0.45 bps per trade is charged
with a monthly trading volume above 30 bnGBP. The minimum fee per trade is 25 pence.
However, on 1 September 2009 the LSE switched back to their traditional fee schedule
with the same pricing scheme for both sides of the market. Chi-X features a maker-taker
pricing scheme with a rebate of 0.20 bps for passive orders and a fee of 0.30 bps for aggressive orders. The LSE and Chi-X feature dynamic tick sizes based on the price of a specific
stock. Since tick size is found to have an influence on market characteristics (cf. Harris,
1994; Goldstein and Kavajecz, 2000; Jones and Lipson, 2001; Bessembinder, 2003b), I control for tick size differences between the LSE and Chi-X in the regression framework. Such
changes, however, only influence a minor part of the sample (13 firms) for a very limited
period of time with the longest time period being 31 trading days. Both markets’ continuous trading sessions start at 8:00 a.m. GMT and last until 16:30 p.m. GMT which is
equivalent to most major continental European exchanges which start at 9:00 a.m. GMT-1
and stop trading at 17:30 p.m. GMT-1.

3.4 Data and Sample Selection
3.4.1 Stock Market Data
Again, trade and quote data are retrieved from the Thomson Reuters DataScope Tick History archive through SIRCA for both, the LSE and the multilateral trading facility Chi-X.
I specifically retrieve trade prices, volumes, and best bid and ask including associated volumes from 1 December 2009 to 31 December 2009. Data entries also include qualifying
codes to identify special trades and quotes. Trades and quotes are timestamped to the millisecond. All prices in the data are reported in British pence. I restrict the analysis to
continuous trading and delete the first and last fiveteen minutes of a trading day. Cutting
the first and last fiveteen minutes avoids biases associated with opening and closing procedures. I also delete all crossing trades from the data. Trades within the spread at the LSE
are also deleted prior to the introduction of hidden liquidity at the LSE. However, those

3 Fragmented Markets and Public Information

68

trades only constitute 0.5% of all trades and regression results do not change if those trades
are left in the data. Table B.3 in Appendix B depicts a sample of raw trade and quote data.
The LSE and Chi-X have individual order books which I retrieve both. For additional
analyses, a consolidated order book is constructed that includes all quotes and trades from
both markets which are then matched based on the Reuters Instrument Code and timestamps. The construction of the consolidated order book is explained in more detail in
Section 3.5. Thomson Reuters also provides an xbo-data stream, a consolidated European
data feed, that merges data of all regulated trading in FTSE 100 stocks. Since this chapter
focuses on the LSE and Chi-X, I construct my own consolidated order book which also
allows for an easy attribution of data entries to either the LSE or Chi-X. In a test with
the four major markets in FTSE 100 trading (LSE, Chi-X, Turquoise, and BATS), only
marginal differences between the constructed consolidated order book and the Thomson
Reuters xbo-stream are found. The analysis focuses on the LSE and Chi-X. With a combined market share in FTSE 100 trading of approximately 85% during 2009, they account
for the major share of trading in those firms. Also, Riordan et al. (2010a) find that the LSE
and Chi-X contribute the major share to price formation.

3.4.2 News Data
This chapter’s analysis is based on the same news data as presented in Chapter 2 Section 2.4,
the news data which are also used throughout this thesis. This chapter specifically uses
news data for firms listed on the LSE. It is important to recall that one news item is scored
separately for different firms and its sentiment can either be positive, negative, or neutral.
A news message that is positive for Vodafone could be negative for British Telecom (both
firms compete in the telecommunications sector) while it might be much more relevant
for British Telecom than for Vodafone. Imagine both companies bid for a large contract
which is eventually awarded to one company. News about this is clearly positive for the
company that won the contract and clearly negative for the other one. Table 3.1 depicts
one sample RNSE message for the Royal Bank of Scotland, a company listed on the LSE
and also the one in the sample with the most negative messages.
The analysis in this chapter relies on the RNSE sentiment measure which is either 1
for positive, -1 for negative, and 0 for neutral news messages. Through the sentiment
measure, I derive information about the average daily general tone of public information
that arrives at trading desks. Since I am interested how the stock specific sentiment of

3.4 Data and Sample Selection

69

public information influences the price discovery process, per firm public information
dummies are constructed which can be used in a regression on a daily basis. To construct
the public information variables for one specific trading day and firm, I aggregate all news
from the end of trading of the last trading day to the end of trading of the current trading
day for which the variable is constructed. In the case of weekends, the variable contains
news from a hypothetical end of trading on Sundays to the end of trading on Mondays. If,
during an aggregation period, no news messages arrive for a specific firm and day, a neutral
sentiment is assigned to this firm/day combination. If the aggregated sentiment is above
zero, this day is considered a day with positive public information for the specific firm, if it
is below zero it is considered to be a negative day, and if it is zero a neutral daily sentiment
is assigned to the respective firm and trading day.

3.4.3 Sample Selection
The sample is based on FTSE 100 constituents which continuously trade in the index over
the year 2009. The FTSE 100 is the most important British stock market also including
the British blue chips. The FTSE 100 represents 85.67% of UK market capitalization as of
31 March 201015 which results in a net market cap of 1,460,100 mGBP. All constituents are
traded on the LSE as well as on Chi-X and represent a broad cross-section of industries.
Stocks in the index are free-float weighted to represent the publicly tradable investment
opportunities. For this analysis, all stocks that are not continuously in the FTSE 100
index during 2009 and stocks which have less than ten trades on one day during 2009 on
either the LSE or Chi-X are removed. Only two stocks are affected by the ‘ten trade rule’.
Additionally, I exclude 24 December 2009 and 31 December 2009 from the data since very
little trading on those days results in extreme values for some measures. The final sample
consists of 88 liquid stocks and 251 trading days in 2009. A complete list of sample firms
including average market capitalization in 2009 can be found in Appendix A.

15

FTSE 100 Index Factsheet,
http://www.ftse.com/Indices/UK_Indices/Downloads/FTSE_100_Index_Factsheet.pdf/.

3 Fragmented Markets and Public Information

70

3.5 Measures
3.5.1 Spreads and Trading Activity
Spread measures are calculated on tick-by-tick data to assess liquidity and also calculate
measures for trading activity. Those measures are then aggregated to a daily frequency per
firm for the regression analysis to capture the intraday market microstructure dynamics of
each variable but to facilitate a comparison of different trading venues. Again, the standard
Lee and Ready (1991) algorithm is used to sign trades with contemporaneous quotes as
proposed by Bessembinder (2003a).
Quote based, ex-ante observable, liquidity is measured with the relative quoted half
spread based on Bessembinder and Kaufman (1997) as described in Chapter 2 Section 2.5.
This measure is based on a quote-to-quote process which is then aggregated to daily per
firm and market averages for estimation purposes. Quoted spreads are also calculated as
quoted spreads at trades for which I need the trade process. Those quoted spreads capture
liquidity represented through the best bid and ask at the time of trades. Quoted spreads
also influence how traders use market or limit orders. When spreads are narrow traders
tend to use market orders while wide spreads incentivize the use of limit orders (Biais et al.,
1995). The effective spread, the spread paid when an incoming market order trades against
a limit order in the order book, is also calculated equivalently to Chapter 2 Section 2.5.
The effective spread additionally captures institutional features of a market such as hidden liquidity through e.g. iceberg orders or real hidden orders and market depth. Hidden
liquidity is available on the LSE from 14 December 2009 and on Chi-X over the entire
observation period. Effective spreads are usually equal to or smaller than quoted spreads
at trade time. If, however, markets feature hidden liquidity inside the spread, effective
spreads might actually be smaller than quoted spreads at trades. For aggregate measures
the relation between quoted spreads at trades and effective spreads fundamentally relies on
the amount of visible and hidden liquidity as well as on other potential institutional details
that might provide price improvement.
In addition to the liquidity measures quoted and effective spread which only include
contemporaneous measures, I compute realized spreads and simple price impacts. The
realized spread measures liquidity suppliers’ revenues independent of the adverse selection
costs imposed on the uninformed by the informed (Bessembinder and Kaufman, 1997).
Let p t be the execution price, D t the trade direction, b t the best bid, a t the best ask then

3.5 Measures

71

the realized spread is calculated with the midpoint five minutes after a trade (z = 5)16 and
the midpoint fifteen minutes after a trade (z = 15) as follows:
r s tz = D t ×

p t − (a t +z + b t +z )/2
(a t + b t )/2

× 10, 000

Price impact is an approximate measure of the adverse selection component of the spread.
The price impact is the effective spread minus the realized spread and tries to measure the
information content of a trade. It approximates the permanent impact of a trade under
the assumption that information impacts are permanent and realized at the 5-minute or
15-minute mark whereas other effects such as inventory costs are transitory. Following
a trade, liquidity suppliers adjust their beliefs about the fundamental value of an asset
depending on the information content of a trade (cf. Glosten and Milgrom, 1985). Using
the same variable definitions as for the measures above, the simple price impact of a trade,
pi tz , is calculated as follows:
p i tz = D t ×

(a t +z + b t +z )/2 − (a t + b t )/2
(a t + b t )/2

× 10, 000

The price impact provides an indication of the information content of a trade. I apply
more robust information measures, not dependent on the spread decomposition, in the
following (cf. Section 3.5.3). Spreads are calculated on both the LSE’s and Chi-X’s individual orderbooks and the consolidated orderbook. I also derive the trade based spread
measures, effective spread, realized spread, and price impact, separated by different trade
size categories. It is differentiated between five trade size categories measured by the number of shares traded17 : 0-499 shares, 500-1,999, 2,000-4,999, 5,000-9,999 shares, and trades
with 10,000 shares traded or more.
To assess trading activity, I calculate for the LSE and for Chi-X the number of trades
per firm and day (#Trades), the number of shares traded per firm and day (Quantity), and
the traded volume per firm and day in GPB (Volume). Comparable to spread measures on
different trade size categories, also the number of trades per category is obtained. Based on
daily per firm volumes the LSE and Chi-X market shares (MktShare) are computed relative
to each other.
16
17

The SEC uses the five minute mark in its definition of realized spreads (Regulation NMS, Rule 605).
Trade size categories are based on SEC trade size categories (Regulation NMS, Rule 600).

3 Fragmented Markets and Public Information

72

Order Book 1
Price
Size
51.00 100
50.00 500
48.50
48.00

300
200

Order Book 2
Price
Size
50.00 200
49.50 300
48.50
47.50

Consolidated Order Book
Price
Size
51.00
100 = 100 + 0
50.00
700 = 500 + 200
49.50
300 = 0 + 300
48.50
48.00
47.50

400 = 300 + 100
200 = 200 + 0
200 = 0 + 200

100
200

Figure 3.1: Consolidated Order Book. Figure 3.1 shows an example of how individual order
books are merged into one consolidated order book.

Spread measures are calculated for the individual order books of the LSE and Chi-X as
well as for the consolidated order book. The individual order books consist of the quotes
and trades of one trading venue, the LSE or Chi-X. The consolidated order book combines
quotes and trades from both trading venues. The best bid or ask is taken from whatever
order book provides the best prices. It might be that the best spread is only provided by
one trading venue or that the bid is provided by one while the ask is provided by the other
trading venue. Figure 3.1 graphically explains how two individual order books are combined to one consolidated order book. Trade based measures can either be calculated in the
consolidated book for all trades not considering the specific trading venue or individually
for both trading venues but with the consolidated order book as their reference. All spread
measures are winsorized at 1% and 99% to account for potential extreme values through
technical data recording errors.

3.5 Measures

73

3.5.2 Information Shares
To measure which market leads in quote based price discovery and in particular how this
characteristic changes on days with high levels of firm specific public information, I compute Hasbrouck (1995) information shares (InfoShares) for each firm and day. The information shares measure assumes the existence of a common efficient price and provides
information on the allocation of price discovery across markets. Joel Hasbrouck specifically intents this measure for fragemented market environments since “fragmentation, the
dispersal of trading in a security in multiple sites, has emerged as a dominant institutional
trend” (Hasbrouck, 1995). The assumption of a common efficient price across markets
implies that stock prices are linked by arbitrage relationships (Hasbrouck, 1995). The information shares measure is based on the concept of cointegration of prices in multiple
markets for one security. “Cointegration refers to the feature that while two price series
[...] may be nonstationary, they do not diverge without bound from each other” (Hasbrouck, 1995). In principal, the econometric model for information shares attemps to
determine which trading venue ‘moves first’.
Econometrically, the price difference between a security trading in two markets is
covariance stationary as a result of arbitrage relationships. The information share attributable to a trading venue is defined as “the proportion of the efficient price innovation
variance that can be attributed to a market” (Hasbrouck, 1995). I use prevailing midpoints
m t of the consolidated order book, based on Thomson Reuters DataScope Tick History
data, assumed to follow a random walk
m t = m t −1 + u t
to characterize the implicit efficient price. Then u t follows a white noise process satisfying
the following criteria: E(u t ) = 0, E(u t2 ) = σ 2 , and E(u t u s ) = 0 for t 6= s. The prices on the
j

LSE and on Chi-X where p t refers to the same security can be written as a vector defined
as follows:
pt =

p tLS E
p C h i -X

!

t

Using above definitions, p t can be written as
pt =

mt +
mt

E
εLS
t
+ εCt h i -X

!
.

3 Fragmented Markets and Public Information

74

The vector of price changes ∆ p t is covariance stationary and may thus be written in a
vector moving average (VMA) representation
∆ p t = ε t + ψ1 ε t −1 + ψ2 ε t −2 + ... = ψ(L)ε t

(3.1)

where ψ(L) is a polynomial in the lag operater and
E
εLS
t
εC h i -X

εt =

!

t

is a vector innovation process with E(ε t ) = 0 and its covariance matrix V a r (ε t ) = Ω. The
E
components of ε t reflect the new information that is incorporated on either the LSE (εLS
)
t
C hi -X
or Chi-X (ε
). According to Huang (2002), Equation 3.1 can be rewritten as
t



∞
∞
X
X


∆ p t = ψ(1)ε t + (1 − L)
ψ j  Li ε t
−
i=1

where L are lag operators, ψ(1) = (In +

P∞

i =1

(3.2)

j =i+1

ψi ), and In is the identity matrix. ψ(1)

“constitutes the long-run impact of a disturbance on each of the prices” (Hasbrouck, 1995)
since it comprises of the sum of all moving average coefficients (Huang, 2002). Based on
Hasbrouck (1995) and Stock and Watson (1988) Equation 3.2 can be represented as follows:
p t = p0 + ψ(1)

t
X

εi + ψ ∗ (L)ε t

i =1

where ψ ∗ (L) is a matrix polynomial in the lag operator. “If price innovations are due to
Pt
new information, the term ψ(1) i=1
εi [greek symbols adjusted] captures the permanent
impact of new information on prices” (Huang, 2002).
Observed midpoint prices are decomposed into a random walk common to all prices
and a covariance stationary error. Based on above definitions, the variance of the random
walk component, the representation of total price discovery, is
σ u2 = ψΩψ0
where ψ is an arbitrary row from ψ(1). In my specific case the variance of the random

3.5 Measures

75

walk reflects price discovery contributions from both trading venues, the LSE and Chi-X:
σ u2 = ψLS E , ψC hi -X
€

2
σLS
E

σLS E,C h i -X

σC h i -X ,LS E

σC2 h i -X

Š

!

ψLS E
ψC h i -X

!

A diagonal covariance matrix Ω (i.e. σi2 = 0) identifies the contribution of each individual
trading venue without ambiguity. The fraction of the variance of a trading venue in relation to the entire variance of the random walk component then provides a measure of a
market’s contribution to price discovery. Formally, this fraction called information shares
is defined by Hasbrouck (1995) as follows:
IS =
j

ψ2j Ω j j
ψΩψ0

where in this chapter j ∈ {LS E, C hi-X }. In a real market setting it might happen that
contemporaneous midpoints of different trading venues are equal. As a result, midpoint
prices may be correlated and Ω is not diagonal. Following Hasbrouck (1995), I determine
upper and lower bounds for each trading venue through maximizing and minimizing the
contribution of each to price discovery. Then the mean contribution of price discovery is
calculated for both, the LSE and Chi-X:
IS

j ,mean

=

IS j ,u p p e r + IS j ,l ow e r
2

Hasbrouck (1995) also proposes to “shorten the interval of observation”. However, my
data is already on a millisecond level and I need to resort to computing upper and lower
bounds. In this chapter, information shares are calculated on a daily per firm basis. Also,
information shares sum up to one by construction.

3.5.3 Trade and Quote Correlated Information
Changes in the efficient price are separated into trade and quote correlated components
differentiating between trades executed on the LSE and Chi-X, j ∈ {LS E, C hi-X }, as
in Hasbrouck (1991a,b). This results in a three-way vector autoregressive (VAR) model.
j

Let x t be the trade direction (-1 for a sell, 1 for a buy) for trades on the LSE or Chi-X,
respectively, and 0 if the trade is not executed on the specific trading venue. r t denotes the

3 Fragmented Markets and Public Information

76

quote midpoint changes in the consolidated order book then the full model is defined as
follows:
rt =

10
X
i=1

x tLS E =

α tr−i r t −i +

10
X

βir r t −i +

i=1
10
X

x tC hi -X =

10
X
i=0

10
X

E LS E
x t −i +
βLS
i

i=1
10
X

γir r t −i +

i=1

αiLS E x tLS−iE +

i =1

10
X
i=0

10
X

γiLS E x tLS−iE +

βCi h i -X x tC−ih i -X + u2,t

i=1
10
X
i=1

αCi h i -X x tC−ih i -X + u1,t

γiC h i -X x tC−ih i -X + u3,t

Then, the VAR model in its vector moving average (VMA) representation is as follows
where L are lag polynomial operators:

 

r
LS E
C h i -X
a
(L)
a
(L)
a
(L)
r

 t  

 LS E   r
LS E
C h i -X
=
  b (L) b (L) b
 xt
(L) 
 


c r (L) c LS E (L) c C h i -X (L)
x tC hi -X
According to Hasbrouck (1991b) the sums of

P∞

a LS E and
t =0

P∞



u
 1,t 


 u2,t 


u3,t

t =0

a C h i -X are used to derive

the cumulative impulse response functions (CIRF) for each trading venue. The CIRF is the
permanent price impact of a trade and it is generally interpreted as the private information
content of a trade. This measure provides for a more precise analysis of information than
the simple price impact of Section 3.5.1. It represents the unexpected part of a trade, the
trade innovation. Trades may contain information at lower frequencies than measured.
However, this measure is used in other studies with the same interpretation (Barclay and
Hendershott, 2003; Madhavan, 2000; Hendershott and Riordan, 2009).
Using the VMA representation from above, information can be decomposed into a trade
correlated part for each trading venue and quote correlated portions (Hasbrouck, 1991b).
The variance decomposition is as follows:
σv2 =

∞
X
i =0

!2
air

σ r2 +

∞
X
i=0

!2
aiLS E

σ x2LS E +

∞
X
i=0

aiC h i -X

!2
σ 2C h i -X
x

The information content of quotes (ICQuote) is the first term, the trade correlated
portions for LSE the second (ICTradeLS E ), and the trade correlated part for Chi-X

3.6 Results and Interpretation

77

(ICTradeC hi -X ) the third. All lags are summed up to get the total trade correlated contribution of each market to price discovery. The results are reported in basis points for the
CIRF and in percent for the information content of trades and quotes. The estimation is
restarted for each trading day and firm in the sample.
j

To approximate the total contribution to price discovery TPDi,d of the LSE and Chi-X,
I compute a combination of information shares and the variance decomposition. Information shares give the percentage that each market contributes to quote based price discovery
whereas the variance decomposition provides the fraction of quote based contribution to
price discovery and trade based contribution for each market separately. Then the following formula for the total contribution of market j for one stock on one day to price
discovery emerges:
TPD j = ICQuote × IS j + ICTrade j .

3.6 Results and Interpretation
3.6.1 Descriptive Statistics
Before I analyze the influence of public information on trading, descriptive statistics on
both markets and the current status of market fragmentation in FTSE 100 trading are presented. Market measures to assess market quality comprise of liquidity measures, trading
activity measures, and information measures on a daily per instrument basis. Additionally,
the trading venues’ market shares are presented. I compute liquidity for both the individual
order books of the LSE and Chi-X and for the consolidated order book. Table 3.2 reports
liquidity measure descriptives. The difference between LSE and Chi-X measures is tested
using clustered standard errors (cf. Petersen, 2009; Thompson, 2011). The most basic liquidity measure, the quoted spread, is 6.08 bps for the incumbent LSE and with 6.80 bps
slightly higher for the MTF Chi-X. Not suprisingly, the consolidated order book provides
a better average spread than the individual order books. The difference between the LSE
and Chi-X is highly significant with a t-value of -8.86. I also compute the quoted spread on
a trade-by-trade basis. Both markets have lower quoted spreads at trades than during periods without trades. This provides evidence in comparison to the overall quoted spreads
that traders monitor the order books of both markets and trade when it is comparatively
cheap to trade. The savings from monitoring the spreads seem to offset monitoring costs.

3 Fragmented Markets and Public Information

78

The difference for quoted spreads at trades between the LSE and Chi-X is small with -0.21
bps but still significant at the 1% level. The same pattern as for quoted spreads emerges for
effective spreads which are slightly smaller by also 0.21 bps on the LSE than on Chi-X. In
the consolidated order book, effective spreads are not significantly different between both
markets, however this changes when effective spreads are splitted by trade size categories.
When differences are statistically significant, it is always the LSE which has smaller effective spreads. Effective spreads are only marginally larger than quoted spreads at trades. The
small difference is an indication that the majority of volume is executed at the best bid or
ask which in turn shows that both markets are highly liquid in FTSE 100 trading. Summarized, the LSE quotes tighter spreads than Chi-X and provides more liquidity measured by
ex-ante and ex-post liquidity measures. Realized spreads are negative on both the LSE and
Chi-X which could be a result of the maker-taker pricing schemes, at least on Chi-X. I find
positive price impacts at the five and fifteen minute marks. However, price impact measures for both markets are not statistically significantly different. Price impacts calculated
on the consolidated order book are significantly higher for the LSE. Since price impact is
a noisy information measure, I use more robust measures in the following paragraphs to
analyze the price discovery process.
Panel B of Table 3.2 and Table 3.3 provide information on trade based spread measures
for different trade sizes. Even for large trade sizes, effective spreads are quite small which
indicates again a highly liquid market. For large trade sizes, the computed difference might
be different to the difference of aggregated individual values. Since large trades do not exist
for all firms and days, the difference can only be calculated when daily values for both
markets exist. Realized spreads at both, the fifteen and the five minute marks, increase
with larger trades sizes both on the LSE and on Chi-X. For the largest trade size, realized
spreads at the 5-minute mark, liquidity suppliers’ revenue, are close to zero at the LSE.
During the observation period the LSE has a relative market share of 73.70% measured
both in volume traded and the number of shares traded. Chi-X attracts on average 26.30%
of the order flow. The relative market share is the market share calculated against the consolidated volume or number of shares traded on the LSE and Chi-X, excluding other MTFs
and OTC trading. The LSE and Chi-X have a combined market share of approximately
85% during 2009 in non-OTC trading. Market shares and other trading activity measures
are reported in Table 3.4. The higher market share of the LSE is attended by more executed
trades, higher volume and a higher number of shares traded on the LSE than on Chi-X per
day and firm. It is interesting that the average trade size is much smaller on Chi-X than on

3.6 Results and Interpretation

79

the LSE with a high statistical significance at the 1% level. The average trade size in GBP is
10,276 on the LSE as compared to 6,160 GBP on Chi-X. One potential explanation could
be that algorithmic traders are more likely to trade on Chi-X since it caters their need for
low latency trading and small round trip times. Algorithmic traders often split orders extensively and use limit order strategies which could result in much smaller trade sizes (cf.
Hendershott and Riordan, 2009; Hendershott et al., 2011). However, the data set in this
chapter does not allow for a verification of this approach. Panel B of Table 3.4 reports the
number of shares traded by trade size categories.
Existing evidence on the influence of market fragmentation on the price discovery process is mixed. Pagano (1989) finds that fragmentation has a negative effect on price discovery whereas Foucault and Menkveld (2008) argue that increased competition through
fragmentation might lead to a deeper consolidated book and thus enhance the price discovery process. Table 3.5 provides descriptives on the price discovery process with measures
calculated on the consolidated order book as presented in Section 3.5.2 and 3.5.3. My analysis focuses on the LSE and Chi-X. Since Riordan et al. (2010a) report that most of the
price discovery in FTSE 100 stocks can be attributed to the LSE and Chi-X, this should
not distort the results. The overall fraction of quote based price discovery is 45.18% with
the remaining information being impounded through trades on the LSE and on Chi-X.
36.54% of total price discovery can be attributed to trades on the LSE and 18.28% to trades
on Chi-X. Chowdhry and Nanda (1991) find that informed trading gravitates to the most
liquid market, which is the LSE over the observaton period in this sample. This is a natural
explanation given that informed traders try to reduce their impact on prices to a minimum
and informed traders are also more likely to use market or marketable limit orders. However, information shares in Panel B of Table 3.5 show that Chi-X leads in quote based price
discovery. 58.19% of quote based price discovery is attributable to Chi-X and only 41.81%
to the LSE. The permanent impact of trade innovation is a proxy for private information impounded into markets through trades. The trade innovation results in Panel A of
Table 3.5 illustrate that much more information, measured in basis points, is impounded
through order flow on the LSE than on Chi-X. Since the LSE is more liquid than Chi-X,
this cannot be a result of low liquidity. Panel C of Table 3.5 reports my measure of total
contribution to price discovery computed through a combination of information shares
and the variance decomposition. The LSE contributes 55.56% to total price discovery and
Chi-X 44.44%.
The results show that price formation takes place on both the LSE and Chi-X and Chi-X

3 Fragmented Markets and Public Information

80

also contributes significant liquidity to trading in FTSE 100 stocks. The further analyses
focus on the impact of firm specific public information on characteristics in the individual
markets and on characteristics of fragmentation.

3.6.2 The Effect of Public Information
News Descriptives
Figure 3.2 shows the number of news messages per month for the year 2009. The number
of news per month is comparably steady with a peak in October and its minimum in
December. Since December is a month with more holidays than the other months of
the year, this result is not suprising. Figure 3.3 depicts the number of news messages per
associated day of the week. Consistent with Berry and Howe (1994), a bit more news
messages arrive Tuesday through Thursday than on Monday and Friday. The graph shows
that neutral news messages amount for the least number of messages and negative messages
for the most news messages. However, differences in the number of news messages for
negative, positive, and neutral news are not dramatic.
Table 3.6 reports descriptive statistics on raw news messages and on the computed per
day and firm public information sentiment. The average sentiment of news during 2009
is negatively skewed with the strongest negative average sentiment in January 2009 and
an almost neutral average sentiment for the third quarter of 2009. 3.56 news messages
arrive for an average firm per day, 1.23 positive messages, 1.60 negative, and 0.73 neutral
items. Overall, the analysis in this chapter comprises 81,507 news messages with an average
sentiment of -0.10. Panel B in Table 3.6 provides information about the calculated per firm
and day public information variables. On average a firm has 45.49 positive, 57.50 negative,
and 148.01 neutral days out of 251 trading days in 2009. The firm with the most positive
days, 125, is Vodafone and the Royal Bank of Scotland features 179 days with on average
negative public information which makes it the firm with the most negative days in 2009
probably as a result of the financial crisis.
Regression Model
To analyze the impact of the tone of public information on trading, I resort to a regression
j

model. Let mi ,d denote all calculated measures on liquidity, trading, and information for
stock i, day d , and market j if applicable. PosNi,d and NegNi,d are dummy variables

3.6 Results and Interpretation

81

10,000
9,000
8,000
7,000
6,000
5,000
4,000
3,000
2,000
1,000
0

Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct Nov Dec

Figure 3.2: News Per Month. Figure 3.2 shows the number of news messages per month after
initial cleaning and data preparation procedures.

per firm and day. They take 1 if a trading day is positive or negative respectively and 0
otherwise. Their coefficients tell whether characteristics of trade and fragmentation move
with the tone of public information. TickLS E and TickC h i -X control for differences in the
tick size between the LSE and Chi-X.

i,d
E
TickLS
i,d

i,d

takes one if the tick size for a specific stock
and day is larger at the LSE than on Chi-X and TickCi ,dh i -X takes 1 in case the tick size is
larger on Chi-X. If tick sizes are equal between both markets both variables take 0. I
include monthly dummy variables to control for time trends and to additionally capture
changes in the maker-taker pricing scheme of the LSE as of 1 September 2009. Then the
following regression model emerges:
E
mi ,d = ai + p1 PosNi ,d + p2 NegNi,d + c1 TickLS
+ c2 TickCi,dh i -X +
i,d
j

11
X
m=1

j

k m Month m + ei,t (3.3)

3 Fragmented Markets and Public Information

82

20,000
18,000
16,000
14,000
12,000
neutral
negative
positive

10,000
8,000
6,000
4,000
2,000
0

Mon

Tue

Wed

Thu

Fri

Figure 3.3: News Per Associated Trading Day of the Week. Figure 3.3 shows the number of
news messages per associated trading day of the week after initial cleaning and data
preparation procedures.

The model is estimated with firm-fixed effects and clustered standard errors (cf. Petersen,
2009; Thompson, 2011). I also test day of the week dummies, those are however insignificant and left out in the final model.
Liquidity and Trading Intensity
One of the most important aspects of financial markets is liquidity. Liquidity allows market participants to trade large sizes at any time at low implicit trading costs. As seen above,
both, the LSE and Chi-X, are highly liquid in trading of FTSE 100 stocks. However, it remains unclear whether and how liquidity changes on individual markets and between both
markets at times of high levels of firm specific public information. Table 3.7 reports results
on different spread measures for the individual order books and for differences between the
LSE and Chi-X for both the positive and negative public information coefficients (PosNi,d
and NegNi ,d ) from Regression 3.3.

3.6 Results and Interpretation

83

Table 3.1: Sample News. Table 3.1 shows one intraday RNSE news message for the firm ‘Royal
Bank of Scotland’ (RBS.L).
Sample RNSE News Item
timestamp
bcast_ref
stock_ric
item_id
relevance
sentiment
sent_pos
sent_neut
sent_neg
lnkd_cnt1
lnkd_cnt2
lnkd_cnt3
lnkd_cnt4
lnkd_cnt5
lnkd_id1
lnkd_id2
lnkd_id3
lnkd_id4
lnkd_id5
lnkd_idpv1
lnkd_idpv2
lnkd_idpv3
lnkd_idpv4
lnkd_idpv5
item_type
item_genre
bcast_text
dsply_name
pnac
story_type
cross_ref
proc_date
take_time
story_date
story_time
named_item
take_seqno
attribtn
prod_code
topic_code
co_ids
lang_ind

20 AUG 2009 16:12:22.554
RBS.L
RBS.L
2009-08-20_16.12.22.nN20424640.T1.bb216780
0.154303
-1
0.0558151
0.12554
0.818645
0
0
0
0
0
.
.
.
.
.
.
.
.
.
.
ARTICLE
NOT DEFINED
Fitch cuts European bank hybrid debt ratings
2
nN20424640
S
.
20-AUG-2009
16:12:22
20-AUG-2009
16:12:22
.
1
RTRS
E U NAW D T NAT M PSC RNP DNP PTD EMK
EUB EUROPE AAA LOA HYD BNK FIN DFIN GB
INS NL DBT CDM FINS WEU LEN RTRS
LLOY.L RBS.L ING.AS SR.AS
EN

3 Fragmented Markets and Public Information

84

Table 3.2: Descriptive Statistics Spreads. Panel A of Table 3.2 provides descriptive statistics for
spread measures. Spread measures comprise quoted spreads, quoted spreads at trades, effective
spreads, realized spreads, and price impacts. Realized spreads are computed both as 5-minute and
15-minute measures. All measures are first aggregated on a daily basis per firm, then tested and
aggregated to overall averages. Table 3.2 presents values for the individual order books of the LSE
and Chi-X as well as the differences (Diff) between the LSE and Chi-X in individual order books
(measureLS E -measureC hi -X ). In addition, values for the consolidated order book are reported for
the LSE (LSE Cons) and Chi-X (Chi-X Cons) individually, for their differences within the consolidated order book (Diff Cons), and for values for the overall consolidated book (Cons). Panel B of
Table 3.2 reports descriptives on effective spreads by trade size measured in the number of shares
traded. Spreads are measured in basis points (bps). Robust t-statistics for the significance of the
differences between the LSE and Chi-X are also reported. Standard deviations over per day and
firm measures are reported in parantheses. ‘a’ indicates significance at the 1% level and ‘b’ indicates
significance at the 5% level.
Measure

Panel A: Spread Measures
LSE

Chi-X

Diff

t-stat
-8.86

LSE Cons

Chi-X Cons

Diff Cons

t-stat

Cons

Quoted Spread

6.08
(3.19)

6.80
(4.82)

-0.72a

Quoted Spread at Trades

4.62
(2.25)

4.83
(2.68)

-0.21a

-4.79

3.66
(1.92)

3.81
(1.93)

-0.15a

-10.30

3.72
(1.91)

Effective Spread

4.69
(2.29)

4.90
(2.74)

-0.21a

-4.81

3.99
(2.04)

4.02
(2.06)

-0.03

-1.76

4.00
(2.03)

Realized Spread 5

-0.57
(1.80)

-0.33
(1.93)

-0.24a

-6.48

-0.57
(1.77)

-0.31
(1.88)

-0.26a

-7.02

-0.47
(1.57)

Realized Spread 15

-0.20
(2.72)

-0.03
(2.97)

-0.16a

-3.96

-0.23
(2.66)

-0.05
(2.88)

-0.18a

-4.30

-0.15
(2.28)

Price Impact 5

5.29
(2.86)

5.26
(3.28)

0.03

0.64

4.60
(2.65)

4.33
(2.63)

0.27a

8.55

4.49
(2.51)

Price Impact 15

4.92
(3.44)

4.97
(3.98)

-0.05

-0.93

4.26
(3.23)

4.07
(3.41)

0.19a

4.96

4.18
(2.94)

Trade Size Category

4.66
(2.56)

Panel B: Effective Spread By Trade Size
LSE

Chi-X

Diff

t-stat

LSE Cons

Chi-X Cons

Diff Cons

t-stat

Cons

-4.96

3.81
(2.05)

3.82
(1.95)

-0.01

-0.29

3.82
(1.98)

< 500

4.52
(2.26)

4.73
(2.70)

-0.21a

500 − 1, 999

4.67
(2.28)

4.97
(2.79)

-0.30a

-5.88

3.96
(2.04)

4.09
(2.08)

-0.13a

-4.81

4.00
(2.02)

2, 000 − 4, 999

4.89
(2.36)

5.36
(3.02)

-0.52a

-8.23

4.20
(2.12)

4.47
(2.36)

-0.32a

-8.87

4.25
(2.10)

5, 000 − 9, 999

5.19
(2.71)

5.62
(3.38)

-0.65a

-9.35

4.46
(2.43)

4.71
(2.74)

-0.45a

-10.39

4.51
(2.41)

≥ 10, 000

6.19
(3.39)

5.83
(3.79)

-0.42a

-5.74

5.34
(3.44)

4.88
(3.18)

-0.22a

-4.68

5.34
(3.41)

3.6 Results and Interpretation

85

Table 3.3: Descriptive Statistics Spreads by Trade Size. Table 3.3 provides descriptive statistics for
realized spreads and price impacts both at the 5 and 15 minute marks by trade size. All measures are
first aggregated to per firm and day values, then tested and aggregated to overall averages. Table 3.3
presents values for the individual order books of the LSE and Chi-X as well as the differences (Diff)
between the LSE and Chi-X in individual order books (measureLS E -measureC h i −X ). Also, values
for the consolidated order book (Cons) are reported not clustered by markets. Panel A of Table 3.3
reports descriptives on realized spreads and Panel B on price impacts. Spreads are measured in basis
points (bps). Robust t-statistics for the significance of the differences between the LSE and Chi-X
are also reported. Standard deviations over per day and firm measures are reported in parantheses.
‘a’ indicates significance at the 1% level and ‘b’ indicates significance at the 5% level.
Panel A
Trade Size Category

Realized Spread 5

Realized Spread 15

LSE

Chi-X

Diff

t-stat

Cons

LSE

Chi-X

Diff

t-stat

Cons

< 500

-0.04
(2.84)

0.03
(3.03)

-0.08

-1.34

0.02
(2.35)

0.18
(4.65)

0.29
(4.73)

-0.10

-1.54

0.24
(3.77)

500 − 1, 999

-0.51
(2.42)

-0.50
(3.12)

-0.01

-0.27

-0.49
(2.06)

-0.16
(4.85)

-0.20
(3.70)

0.04

0.77

-0.17
(3.03)

2, 000 − 4, 999

-1.37
(4.37)

-1.27
(6.72)

-0.02

-0.31

-1.33
(4.12)

-0.86
(6.28)

-0.81
(10.50)

0.06

0.61

-0.88
(5.81)

5, 000 − 9, 999

-2.07
(7.92)

-1.81
(12.09)

0.01

0.08

-2.03
(7.62)

-1.34
(11.90)

-1.06
(18.75)

-0.09

-0.55

-1.37
(11.47)

-0.08
(15.09)

-2.51
(15.56)

1.05a

4.95

-0.18
(14.34)

1.20
(21.75)

-1.69
(23.99)

1.11a

3.21

1.05
(20.81)

≥ 10, 000

Panel B
Trade Size Category

Price Impact 5
LSE

Price Impact 15

Chi-X

Diff

t-stat

Cons

LSE

Chi-X

Diff

t-stat

Cons

-2.22

3.82
(2.77)

4.36
(4.92)

4.48
(5.24)

-0.12

-1.52

3.60
(4.02)

< 500

4.58
(3.28)

4.73
(3.83)

-0.14a

500 − 1, 999

5.21
(3.21)

5.50
(4.18)

-0.30a

-5.00

4.50
(2.73)

4.86
(4.18)

5.21
(5.60)

-0.36a

-4.45

4.19
(3.49)

2, 000 − 4, 999

6.30
(5.02)

6.67
(7.57)

-0.51a

-5.41

5.62
(4.72)

5.80
(6.74)

6.22
(11.07)

-0.59a

-4.74

5.81
(6.25)

5, 000 − 9, 999

7.35
(8.26)

7.51
(12.50)

-0.70a

-5.93

6.76
(7.63)

6.66
(11.98)

6.78
(18.97)

-0.60a

-3.95

6.16
(11.18)

6.81
(14.13)

8.45
(16.07)

-1.37a

-6.55

7.32
(11.71)

5.94
(20.10)

7.65
(24.35)

-1.33a

-4.24

6.72
(17.55)

≥ 10, 000

3 Fragmented Markets and Public Information

86

Table 3.4: Descriptive Statistics Trading Activity. Table 3.4 provides descriptive statistics for
trading activity measures. All measures are first aggregated on a daily basis per firm, then tested and
aggregated to overall averages. Table 3.4 presents values for the individual order books of the LSE
and Chi-X as well as the differences (Diff) between the LSE and Chi-X (measureLS E -measureC h i−X ).
Panel A reports descriptives on the number of trades per day and firm (#Trades), the volume in
kGBP, the quantity, the average trade size in volume, the average trade size by shares traded, and
market shares by volume and quantity. Panel B reports descriptives on the number of trades per
day and firm splitted by trade size categories measured in the number of shares traded. Robust
t-statistics for the significance of the differences between the LSE and Chi-X are reported in the last
column. Standard deviations are reported in parantheses. ‘a’ indicates significance at the 1% level
and ‘b’ indicates significance at the 5% level.
Measures

Panel A: Trading Activity
LSE

Chi-X

Diff

t-stat

2,318
(1,967)

1,346
(1,154)

972a

11.16

Volume (in kGBP) Per Day and Firm

28,852
(40,029)

10,196
(13,378)

18, 656a

7.49

Quantity (in kShares) Per Day and Firm

7,799
(18,706)

2,547
(5,611)

5, 253a

4.71

Trade Size (Volume)

10,276
(5,484)

6,160
(3,251)

4, 117a

19.97

Trade Size (Quantity)

2,723
(3,829)

1,610
(2,292)

1, 114a

7.09

Market Share (Volume)

73.70%
(9.08%)

26.30%
(9.08%)

47.40%a

44.26

Market Share (Quantity)

73.70%
(9.08%)

26.30%
(9.08%)

47.40%a

44.26

Trade Size Category

Panel B: Number of Trades Per Day and Firm

#Trades

LSE

Chi-X

Diff

t-stat
8.24

< 500

725.77
(721.75)

547.13
(589.86)

178.65a

500 − 1, 999

810.85
(709.35)

496.34
(459.38)

314.51a

8.87

2, 000 − 4, 999

433.57
(485.73)

194.57
(276.71)

239.01a

9.98

5, 000 − 9, 999

192.33
(329.08)

67.75
(156.14)

124.58a

6.85

≥ 10, 000

155.74
(474.13)

40.51
(159.16)

115.23a

3.95

3.6 Results and Interpretation

87

Table 3.5: Descriptive Statistics Price Discovery. Table 3.5 provides descriptive statistics for information measures based on Hasbrouck (1991a,b, 1995). All measures are first computed per firm
and day, then aggregated to overall averages and tested. Statistics on the LSE, Chi-X, and the differences between the two markets (Diff) are presented. Panel A reports descriptives on trade based
price discovery: the share of trade based price discovery in total price discovery (% Trade Based)
and the permanent impact of trade innovation in basis points (Trade Innovation). Panel B reports
descriptives on quote based price discovery: the overall share of quote based price discovery in
total price discovery and the share of the LSE and Chi-X respectively in quote based price discovery. Panel C provides information about the total contribution to price discovery of the LSE and
Chi-X. Robust t-statistics for the significance of the differences between the LSE and Chi-X are
also reported. Standard deviations over per day and firm measures are reported in parantheses. ‘a’
indicates significance at the 1% level and ‘b’ indicates significance at the 5% level.
Panel A: Trade Based Price Discovery
% Trade Based

Trade Innovation

Value

StdDev

Value

StdDev

LSE

36.54%

(8.84%)

2.31

(2.11)

Chi-X

18.28%

(12.35%)

1.45

(2.51)

Diff
t-stat

18.25%a
27.33

0.86a
15.16

Panel B: Quote Based Price Discovery
% Quote Based
StdDev

45.18%

(15.59%)

Information Shares
Value

StdDev

LSE

41.81%

(18.63%)

Chi-X

58.19%

(18.63%)

Overall

Value

-16.38%a
-9.36

Diff
t-stat

Panel C: Total Contribution to Price Discovery
Fraction of PD
Value

StdDev

LSE

55.56%

(14.69%)

Chi-X

44.44%

(14.69%)

Diff
t-stat

11.12%a
7.38

3 Fragmented Markets and Public Information

88

Table 3.6: Descriptive Statistics News. The news data are based on the FTSE 100 stock sample
of my analysis with 88 firms. Panel A presents overall statistics on raw messages with standard
deviations in parantheses. Statistics are presented for per day and firm averages, the whole year of
2009, and individual months of 2009. ‘Positive News’, ‘Negative News’ and ‘Neutral News’ report
the raw number of news messages with the respective sentiment. ‘All News’ presents statistics
about news not clustered by sentiment. ‘Avg.Sentiment’ gives the overall average sentiment of
RNSE raw messages. Panel B reports results on the computed firm specific public information
sentiment which is a per day and firm average value. Statistics for the whole year of 2009, each
month, and the most extreme firms are presented. The number of positive, negative, and neutral
days as well as the overall number of trading days per firm are reported.
Panel A: Raw News Messages
Positive News

Negative News

Neutral News

All News

Avg.Sentiment

1.23
(3.09)

1.60
(4.17)

0.73
(4.22)

3.56
(8.65)

-0.08
(0.64)

Year 2009

28,158

36,638

16,711

81,507

-0.10

January
February
March
April
May
June
July
August
September
October
November
December

1,681
2,022
2,107
1,976
2,400
2,406
2,484
2,214
3,015
3,329
2,692
1,832

3,819
3,819
3,721
2,950
3,158
3,132
3,498
2,244
2,339
3,384
2,582
1,992

1,224
1,296
1,269
1,178
1,312
1,461
1,684
1,172
1,220
2,459
1,489
947

6,724
7,137
7,097
6,104
6,870
6,999
7,666
5,630
6,574
9,172
6,763
4,771

-0.32
-0.25
-0.23
-0.16
-0.11
-0.10
-0.13
-0.01
0.10
-0.01
0.02
-0.03

News Per Day
and Firm (Avg.)

Panel B: Per Day and Firm Computed Public Information Sentiment
Positive Days

Negative Days

Neutral Days

All Trading Days

Year 2009

45.49

57.50

148.01

251

January
February
March
April
May
June
July
August
September
October
November
December

2.82
2.69
3.10
3.38
3.49
3.75
4.05
3.55
5.27
4.92
4.88
3.60

5.64
5.41
6.25
4.83
4.89
5.18
5.77
3.80
3.82
4.66
4.08
3.18

12.55
11.19
12.65
11.80
10.63
13.07
13.18
12.66
12.91
12.42
12.05
12.22

21
20
22
20
19
22
23
20
22
22
21
19

Firm with most
positive days (Vodafone)

125

85

41

251

49

179

23

251

Firm with most
negative days (RBS)

-0.01
(-0.60)

-0.00
(-0.19)

Quoted Spread at Trades
Coeff.
t-stat

Effective Spread
Coeff.
t-stat

continued on next page . . .

-0.02
(-0.47)

0.12a
(3.67)

0.11a
(3.43)

0.14a
(2.62)

0.02
(0.51)

0.01
(0.21)

0.04
(0.68)

0.12a
(3.18)

0.11a
(2.94)

0.21 b
(2.51)

Chi-X
pos
neg

-0.02
(-1.52)

-0.02
(-1.60)

-0.05
(-1.51)

0.01
(0.40)

0.01
(0.53)

-0.07
(-1.39)

Diff
pos
neg

Spreads in the Individual Order Books
LSE
pos
neg

Quoted Spread
Coeff.
t-stat

Measure

-0.01
(-0.48)

-0.02
(-0.85)

0.09a
(3.37)

0.07a
(2.91)

LSE Cons
pos
neg

-0.00
(-0.11)

-0.00
(-0.11)

0.08a
(3.11)

0.06 b
(2.49)

Chi-X Cons
pos
neg

-0.01
(-0.85)

0.00
(0.07)

0.01
(1.00)

0.01
(1.50)

Diff Cons
pos
neg

Spreads in the Consolidated Order Book

-0.01
(-0.40)

-0.02
(-0.95)

-0.03
(-1.06)

0.09a
(3.34)

0.07a
(2.83)

0.07
(1.88)

Cons
pos
neg

Table 3.7: Spreads and Public Information. Table 3.7 provides regression (cf. Equation 3.3) results for spread measures: quoted spreads,
quoted spreads at trade time, effective spreads, realized spreads, and price impacts. Realized spreads and price impacts are computed both as
5-minute and 15-minute measures. All measures are first aggregated to daily values per firm, then included in the regression model. The relevant
variables are dummy variables for positive and negative public information days. Table 3.7 presents results for the individual order books of
the LSE and Chi-X as well as the differences (Diff) between the LSE and Chi-X in individual order books of the two markets. I also report
coefficients for regressions with the consolidated order book for the LSE (LSE Cons), Chi-X (Chi-X Cons), the differences of the two markets
in the consolidated order book (Diff Cons), and the consolidated order book without clustering by markets (Cons). The regression model
controls for tick size differences between the LSE and Chi-X, includes month of the year dummy variables, and is estimated with firm-fixed
effects. Spreads are measured in basis points (bps). Robust t-statistics for the significance of the differences between the LSE and Chi-X are
reported in parantheses. ‘a’ indicates significance at the 1% level and ‘b’ indicates significance at the 5% level.

3.6 Results and Interpretation
89

0.05
(1.54)

-0.01
(-0.18)

-0.05
(-1.34)

0.01
(0.13)

Realized Spread 15
Coeff.
t-stat

Price Impact 5
Coeff.
t-stat

Price Impact 15
Coeff.
t-stat
0.21a
(2.68)

0.13 b
(2.42)

-0.08
(-1.40)

-0.01
(-0.23)

0.04
(0.67)

-0.01
(-0.14)

-0.03
(-0.44)

0.02
(0.66)

0.29a
(4.02)

0.19a
(3.31)

-0.17a
(-2.53)

-0.07
(-1.65)

Chi-X
pos
neg

-0.04
(-0.72)

-0.05
(-1.35)

0.02
(0.34)

0.03
(0.83)

-0.08
(-1.39)

-0.05
(-1.31)

0.09
(1.37)

0.06
(1.44)

Diff
pos
neg

Spreads in the Individual Order Books
LSE
pos
neg

Realized Spread 5
Coeff.
t-stat

Measure

. . . continued from Table 3.7

-0.01
(-0.25)

-0.06
(-1.68)

0.00
(0.04)

0.06
(1.59)

0.15 b
(2.24)

0.09
(1.89)

-0.06
(-1.20)

-0.00
(-0.06)

LSE Cons
pos
neg

0.01
(0.24)

-0.03
(-0.81)

-0.02
(-0.33)

0.03
(0.71

0.23a
(3.56)

0.14a
(3.06)

-0.15a
(-2.59)

-0.06
(-1.65)

Chi-X Cons
pos
neg

-0.03
(-0.52)

-0.03
(-0.97)

0.02
(0.40)

0.04
(0.77)

0.01
(0.82)

-0.05
(-1.22)

0.09
(1.39)

0.06
(1.55)

Diff Cons
pos
neg

Spreads in the Consolidated Order Book

0.01
(0.01)

-0.05
(-1.36)

-0.01
(-0.24)

0.04
(1.25)

0.18a
(2.93)

0.11 b
(2.42)

-0.09 b
(-1.97)

-0.02
(-0.68)

Cons
pos
neg

90
3 Fragmented Markets and Public Information

3.6 Results and Interpretation

91

Overall, coefficients for liquidity are statistically significant for negative public information days and insignificant for all measures on positive public information days. Quoted
spreads, quoted spreads at trades, and the ex-post measured effective spreads all increase
on both, the LSE and on Chi-X, when traders receive on average negative public information. On the LSE, quoted spreads increase at the 1% significance level by 0.14 bps, quoted
spreads at trades by 0.11 bps, and effective spreads by 0.12 bps. Quoted spreads increase
at the 5% level by 0.21 bps on Chi-X while quoted spreads at trades increase by 0.11 bps
and effective spreads by 0.12 bps. Although the magnitude of the coefficients suggests
that the reduction in visible liquidity is stronger on Chi-X then on the LSE, I do not find
statistically significant differences between the individual order books.
There is practically no difference in the reduction of liquidity at trade times measured
by the quoted spread at trades and the effective spread. For days with positive public
information, I do not find significant coefficients, neither for the individual order books
nor for the difference between the LSE and Chi-X. The signs of coefficients for quoted
spreads, quoted spreads at trade, and effective spreads are consistently negative for the LSE
and positive for Chi-X, meaning an increase in liquidity at the LSE with a concurrent
decrease in liquidity at Chi-X, but the coefficients are not statistically significant. Spreads
in the consolidated order book react comparably to the individual order books. Quoted
spreads at trades, pooled over both markets, significantly increase by 0.07 bps on negative
days and do not significantly change on positive days. Effective spreads also increase by
0.09 bps on negative public information days which translates into higher execution costs,
also over all trade size categories (Table 3.8) except for trades larger than 10,000 shares in
the consolidated book.
Although, the analysis in Chapter 2 is performed on intraday high-frequency data, my
results are quite consistent with the findings in Chapter 2. I find a significant drop in
liquidity around negative news and an improvement in liquidity around positive news in
Chapter 2 which can be explained with competition for liquidity supply in the limit order
book. Possibly, due to the necessary daily aggregation of data in this chapter which is on a
lower frequency, I do not find significant results for positive news. Overall, in this chapter
liquidity drops on both markets on negative days and does not change significantly on positive days. Liquidity providers might want to protect themselves against better informed
traders or highly capable public information processers in a negative firm specific public
information environment (cf. Kim and Verrecchia, 1994). This argument is supported by
the realized spread results for Chi-X as reported in Table 3.7. Realized spreads, liquidity

3 Fragmented Markets and Public Information

92

suppliers’ revenue, fall at the fifteen minute mark significantly by 0.17 bps for negative
news. In the consolidated book, realized spreads at the fiveteen minute mark fall by 0.09
bps. Price impacts increase statistically significantly for both markets at the five and fifteen minute marks for days with negative firm specific public information. I do not find a
change in price impact for positive days. An increase in price impact hints at more private
information impounded, which fits the realized spread results and a slight observed reduction in liquidity. Differences between the LSE and Chi-X for realized spreads and price
impacts are generally not significant. In the consolidated orderbook, on negative days a
slight decrease in realized spreads, liquidity suppliers revenue, and an increase in price impacts is found. Overall, regressions with spreads calculated on the consolidated order book
show qualitatively the same results as spreads calculated on the individual order books.
Effective spreads, realized spreads, and price impacts by trade size categories, reported in
Tables 3.9, 3.10, and 3.11, paint on average essentially the same picture as the overall measures. One interesting aspect though is that mid-sized trades between 500 and 1,999 shares
which have been found in previous research to be more informed (Barclay and Warner,
1993) need to pay even higher effective spreads on Chi-X than on the LSE on positive public information days than normal (see Table 3.9). The difference is statistically significant
at the 5% level. Over all trading days, the difference for mid-sized trades between the LSE
and Chi-X is already 0.30 bps (1% level, see Table 3.2) with the LSE having an average
effective spread of 4.67 bps and Chi-X one of 4.97 bps. This difference increases by 0.03
bps on days with positive public information. Overall, trade size category 500 to 1,999
has also the largest difference for the absolute number of trades per day of 314.51 trades
(Panel B, Table 3.4). Summerized, the LSE provides lower execution costs for mid-sized
trades and has the higher market share in such trades which potentially have also a high
level of private information.

< 500
Coeff.
t-stat
500 − 1, 999
Coeff.
t-stat
2, 000 − 4, 999
Coeff.
t-stat
5, 000 − 9, 999
Coeff.
t-stat
≥ 10, 000
Coeff.
t-stat

Trade Size Category

0.05
(1.69)
0.03
(0.35)
0.01
(0.10)

0.09a
(3.37)
0.11a
(3.97)
0.11a
(2.98)
0.08
(1.62)

-0.01
(-0.28)
0.01
(0.60)
-0.00
(-0.02)
0.02
(0.40)

-0.08
(-0.24)

0.01
(0.24)

0.06a
(2.59)

-0.34
(-1.11)

-0.19
(-1.48)

0.03
(0.49)

0.02
(0.48)

0.07
(1.47)

RSpread 5
pos
neg

-0.02
(-0.90)

ESpread
pos
neg

-0.00
(-0.01)

0.03
(0.14)

0.04
(0.31)

-0.04
(-1.04)

0.03
(0.34)

-0.17
(-0.43)

-0.23
(-1.14)

-0.07
(-0.75)

-0.01
(-0.19)

0.01
(0.09)

RSpread 15
pos
neg

-0.11
(-0.50)

-0.07
(-0.55)

-0.02
(-0.24)

-0.06
(-1.73)

-0.03
(-0.70)

0.12
(0.55)

0.20
(1.57)

0.08
(1.30)

0.07
(1.50)

-0.01
(-0.19)

PImpact 5
pos
neg

Consolidated Order Book

-0.30
(-0.99)

-0.08
(-0.43)

-0.03
(-0.24)

0.04
(0.78)

-0.05
(-0.66)

-0.22
(-0.65)

0.19
(1.01)

0.17c
(1.89)

0.10
(1.50)

0.06
(0.72)

PImpact 15
pos
neg

Table 3.8: Spreads in the Consolidated Order Book and Public Information. Table 3.8 provides regression results for spread measures by
trade size measured in the number of shares traded. Spread measures comprise effective spreads, realized spreads, and price impacts. Realized
spreads and price impacts are computed both as 5-minute and 15-minute measures. All measures are first aggregated to daily values per firm and
then included in the regression model. The relevant variables are dummy variables for positive and negative public information days. Table 3.8
presents results for the consolidated order book. The regression model controls for tick size differences between the LSE and Chi-X, includes
month of the year dummy variables, and is estimated with firm-fixed effects. Spreads are measured in basis points (bps). Robust t-statistics
for the significance of the differences between the LSE and Chi-X are reported in parantheses. ‘a’ indicates significance at the 1% level and ‘b’
indicates significance at the 5% level.

3.6 Results and Interpretation
93

< 500
Coeff.
t-stat
500 − 1, 999
Coeff.
t-stat
2, 000 − 4, 999
Coeff.
t-stat
5, 000 − 9, 999
Coeff.
t-stat
≥ 10, 000
Coeff.
t-stat

Trade Size Cat.

0.03
(0.88)
0.04
(1.14)
-0.02
(-0.42)

0.12a
(3.63)
0.14a
(4.16)
0.15a
(3.35)
0.09
(1.30)

-0.00
(-0.14)

0.02
(0.64)

-0.01
(-0.29)

-0.02
(-0.34)

0.13
(1.82)

-0.00
(-0.09)

0.09a
(3.05)

0.01
(0.31)
-0.12 b
(-2.01)

0.15 b
(2.52)

-0.03
(-1.59)

-0.03 b
(-1.96)

-0.02
(-1.31)

-0.01
(-0.23)

0.05
(0.89)

-0.04 b
(-2.05)

0.00
(0.19)

0.00
(0.25)

Diff
pos
neg

0.08
(1.65)

0.17a
(4.59)

0.12a
(3.21)

0.09 b
(2.56)

Chi-X
pos
neg

-0.02
(-0.94)

LSE
pos
neg

Individual Order Books

Effective Spread

-0.00
(-0.00)

-0.00
(-0.03)

0.02
(0.66)

-0.01
(-0.57)

-0.02
(-1.06)

0.08
(1.48)

0.11a
(3.00)

0.11a
(3.95)

0.09a
(3.25)

0.07a
(2.66)

LSE
pos
neg

0.09
(1.54)

-0.04
(-1.06)

0.01
(0.49)

-0.00
(-0.21)

-0.02
(-0.81)

0.10
(1.70)

0.03
(0.58)

0.12a
(4.11)

0.09a
(3.43)

0.06 b
(2.34)

Chi-X
pos
neg

-0.09
(-1.58)

0.03
(0.84)

-0.01
(-0.29)

-0.02
(-1.49)

-0.01
(-0.44)

-0.01
(0.22)

0.06
(1.37)

-0.02
(-1.67)

0.00
(0.10)

0.01
(0.82)

Diff
pos
neg

Consolidated Order Book

Table 3.9: Effective Spread by Trade Size and Public Information. Table 3.9 provides regression results for effective spreads by trade size
measured in the number of shares traded. All measures are first aggregated to daily values per firm and then included in the regression model.
The relevant variables are dummy variables for positive and negative public information days. Table 3.9 presents results for the individual
order books of the LSE and Chi-X as well as the differences (Diff) between the LSE and Chi-X in individual order books of the two markets.
I also report coefficients for regressions with the consolidated order book for the LSE (LSE Cons), Chi-X (Chi-X Cons), the differences of
the two markets in the consolidated order book (Diff Cons), and the consolidated order book without clustering by markets (Cons). The
regression model controls for tick size differences between the LSE and Chi-X, includes month of the year dummy variables, and is estimated
with firm-fixed effects. Spreads are measured in basis points (bps). Robust t-statistics for the significance of the differences between the LSE
and Chi-X are reported in parantheses. ‘a’ indicates significance at the 1% level and ‘b’ indicates significance at the 5% level.

94
3 Fragmented Markets and Public Information

< 500
Coeff.
t-stat
500 − 1, 999
Coeff.
t-stat
2, 000 − 4, 999
Coeff.
t-stat
5, 000 − 9, 999
Coeff.
t-stat
≥ 10, 000
Coeff.
t-stat

Trade Size Cat.

0.11 b
(2.00)
0.02
(0.44)
-0.01
(-0.13)
-0.22
(-1.65)
-0.47
(-1.37)

0.05
(0.83)

0.07 b
(2.12)

0.00
(0.05)

0.02
(0.16)

-0.22
(-0.64)

LSE
pos
neg

-0.50
(-1.41)

-0.20
(-0.65)

0.08
(0.51)

-0.00
(-0.08)

-0.04
(-0.77)

-0.21
(-0.51)

-0.34
(-1.34)

0.03
(0.19)

-0.05
(-0.83)

-0.01
(-0.24)

Chi-X
pos
neg

Realized Spread 5

0.22
(0.68)

0.19
(0.61)

-0.02
(-0.15)

0.08
(1.37)

0.09
(1.70)

-0.02
(-0.05)

0.22
(0.79)

-0.00
(-0.02)

0.06
(1.47)

0.12 b
(2.53)

Diff
pos
neg

Individual Order Books

-0.46
(-0.36)

0.02
(0.09)

0.02
(0.13)

-0.03
(-0.74)

0.09
(0.88)

-0.25
(-0.55)

-0.24
(-1.12)

-0.06
(-0.66)

-0.00
(-0.01)

0.10
(0.86)

LSE
pos
neg

-0.37
(-0.80)
-0.85
(-1.27)

-1.41 b
(-2.13)

-0.28
(-1.45)

-0.04
(-0.53)

-0.12
(-1.73)

-0.06
(-0.11)

-0.08
(-0.44)

-0.07
(-1.03)

-0.08
(-0.97)

Chi-X
pos
neg

Realized Spread 15

0.98
(1.54)

0.10
(0.18)

0.16
(0.87)

0.04
(0.53)

0.17
(1.53)

0.54
(0.81)

0.18
(0.37)

0.22
(1.13)

0.04
(0.51)

0.22
(1.77)

Diff
pos
neg

Table 3.10: Realized Spread by Trade Size and Public Information. Table 3.10 provides regression results for realized spreads at the 5and 15-minute marks by trade size measured in the number of shares traded. All measures are first aggregated daily values per firm and then
included in the regression model. The relevant variables are dummy variables for positive and negative public information days. Table 3.10
presents results for the individual order books of the LSE and Chi-X as well as the differences (Diff) between the LSE and Chi-X in individual
order books of the two markets. The regression model controls for tick size differences between the LSE and Chi-X, includes month of the
year dummy variables, and is estimated with firm-fixed effects. Spreads are measured in basis points (bps). Robust t-statistics for the significance
of the differences between the LSE and Chi-X are reported in parantheses. ‘a’ indicates significance at the 1% level and ‘b’ indicates significance
at the 5% level.

3.6 Results and Interpretation
95

< 500
Coeff.
t-stat
500 − 1, 999
Coeff.
t-stat
2, 000 − 4, 999
Coeff.
t-stat
5, 000 − 9, 999
Coeff.
t-stat
≥ 10, 000
Coeff.
t-stat

Trade Size Cat.
LSE
pos
neg

-0.06
(-0.34)
0.13
(0.42)

0.11
(1.92)
0.15 b
(2.18)
0.36 b
(2.37)
0.51
(1.60)

-0.07
(-1.75)

0.02
(0.17)

-0.03
(-0.25)

0.05
(0.19)

0.63
(1.79)

0.03
(0.55)

-0.01
(-0.24)

0.04
(0.78)

0.33
(0.82)

0.39
(1.59)
-0.35
(-1.01)

-0.13
(-0.43)

0.00
(0.02)

-0.10
(-1.95)

0.17a
(2.59)
0.14
(0.96)

-0.11 b
(-2.13)

0.11
(1.66)

0.08
(0.17)

-0.13
(-0.50)

-0.03
(-0.21)

-0.06
(-1.36)

-0.13a
(-2.64)

-0.15
(-0.41)

-0.04
(-0.18)

0.00
(0.01)

0.03
(0.64)

-0.11
(-1.06)

0.16
(0.38)

0.34
(1.52)

0.21
(1.94)

0.12
(1.56)

-0.00
(-0.01)

0.44
(0.97)
0.99
(1.52)

1.53 b
(2.39)

0.45 b
(2.26)

0.16 b
(2.12)

0.22a
(2.60)

0.02
(0.04)

0.11
(0.60)

0.10
(1.34)

0.08
(1.06)

Chi-X
pos
neg

Diff
pos
neg

Chi-X
pos
neg

-0.07
(-1.18)

LSE
pos
neg

Price Impact 15

Price Impact 5

Individual Order Books

-1.12
(-1.83)

-0.08
(-0.14)

-0.18
(-0.98)

-0.06
(-0.96)

-0.19
(-1.74)

-0.50
(-0.76)

-0.13
(-0.27)

-0.26
(-1.30)

-0.03
(-0.45)

-0.22
(-1.84)

Diff
pos
neg

Table 3.11: Price Impact by Trade Size and Public Information. Table 3.11 provides regression results for price impacts at the 5- and 15minute marks by trade size measured in the number of shares traded. All measures are first aggregated to values per firm on a daily basis and
then included in the regression model. The relevant variables are dummy variables for positive and negative public information days. Table 3.11
presents results for the individual order books of the LSE and Chi-X as well as the differences (Diff) between the LSE and Chi-X in individual
order books of the two markets. The regression model controls for tick size differences between the LSE and Chi-X, includes month of the
year dummy variables, and is estimated with firm-fixed effects. Price impacts are measured in basis points (bps). Robust t-statistics for the
significance of the differences between the LSE and Chi-X are reported in parantheses. ‘a’ indicates significance at the 1% level and ‘b’ indicates
significance at the 5% level.

96
3 Fragmented Markets and Public Information

3.6 Results and Interpretation

97

Table 3.12 reports regression results for measures of trading activity. I apply the regression model to both the absolute difference between markets and the relative difference
calculated through ratios. Ratios18 give an impression of the change in trading activity
relative to trading activity on normal days and account for the difference in market shares
between the LSE and Chi-X. Results show a highly significant increase in trading activity on positive as well as on negative public information days measured by the number
of trades per day, the number of shares traded, and volume. I use the natural logarithm
of daily trading volume and quantity for the regression model. The increase in trading
volume demonstrates that investors have differential interpretations of both negative and
positive public information (cf. Kim and Verrecchia, 1991). The increase in trading activity is relatively equal between positive and negative days. For instance, on average the LSE
has 2,318 trades per day and firm, Chi-X has 1,346. Negative days show an increase of 257
trades on the LSE and 106 on Chi-X compared to neutral days. On positive days, I find
an increase of 218 on the LSE and 117 on Chi-X. Looking at volume, quantity, and the
number of trades per day, I find that both measured by differences in trading activity and
ratios, trading activity increases relatively more on the LSE than on Chi-X. The increase
in trading activity is consistent with existing empirical (Ryan and Taffler, 2007; Mitchell
and Mulherin, 1994) and theoretical literature (Kim and Verrecchia, 1991, 1994).
Per trade volume increases strongly on both markets but again statistically significantly
stronger on the LSE than on Chi-X relative to their per trade volume during normal times.
The absolute difference between the LSE and Chi-X as well as the ratio are both statistically significant. One possible explanation could be that high levels of public information
increase differential interpretation among traders which can be seen through an increase
in trading activity. If traders are more differential in their interpretation of certain stocks,
they might want to execute their trades faster than normal and resort to larger trade sizes
despite slightly worse liquidity on negative days and no increase in liquidity on positive
days. It is interesting that the trade size increases stronger on the LSE, the more liquid
market and the market with the a priori significantly larger average trade size (10,276 GBP
on the LSE vs. 6,160 GBP on Chi-X). The LSE market share increases significantly on
both positive and negative days. However, the increase is much stronger with 0.87% on
positive days than with 0.30% on negative days. Since Chi-X’s market share concurrently
decreases by 0.87% on positive days I find a shift of almost 2% in trading of FTSE 100
18

ra t i o =

meas u r e LS E
meas u r e C hi -X

3 Fragmented Markets and Public Information

98
stocks on days with public information.

Panel B of Table 3.12 reports regression results on the number of trades per day and
firm for different trade size categories. For positive public information days, one sees
a significant rise of trading in all but the largest trade size category. On negative days,
trading activity in all trade size categories increases significantly. The largest difference
in the increase can be found in the mid-size trade category (500 – 1,999). The difference
between the LSE and Chi-X increases by 38.16 trades for positive days and 41.64 trades for
negative days both statistically significant at the 1% level. Interestingly, this is also the same
trade size category with a significant change in the difference in effective spreads between
the LSE and Chi-X on positive days (cf. Table 3.9). I find the highest absolute increase
in trading activity by trade size for both the LSE and Chi-X again in the mid-size trade
category. The patterns that are found in mid-sized trades will be examined more closely in
the next section with robust information measures.
Information
To shed light on the question how stock specific markets characteristics and characteristics
of fragmentation vary with firm specific public information, I take a look at the triad of
liquidity, trading activity, and information with the help of robust information measures
(see Section 3.5). Descriptive statistics (see Table 3.5) show that Chi-X contributes more
to quote based price discovery, the LSE more to trade based price discovery and altogether
also more to total price discovery.
Table 3.13 reports results on the price discovery process and public information. During times with high levels of public information, I do not find significant changes in the
fractions of trade based and the fraction of quote based price discovery. On positive public information days, I find a strong decrease in private information on Chi-X of 0.19 bps
and a slight but statistically insignificant decrease on the LSE. Overall, there is less private information impounded into the markets on positive days. Interestingly, the private
information that still is impounded shifts to the LSE away from Chi-X. Additionally, information that translates into prices through quotes also shifts significantly to the LSE
with a difference between the LSE and Chi-X of 1.17%. Both, quote based price discovery and trade based price discovery combined to total price discovery, shift from Chi-X
to the LSE by 1.47% and statistically significant at the 1% level. I find different effects
on days with public information of an on average negative tone. Private information in-

3.6 Results and Interpretation

99

creases on the LSE significantly while there is no statistically significant change on Chi-X.
In combination, comparable to positive days, private information shifts from Chi-X to the
LSE. A general increase in private information as a result of negative news is consistent
with Chapter 2. Neither information shares nor the total contribution to price discovery
change statistically significant on negative days.
Overall, I find that a negative tone of public information decreases liquidity, increases
trading activity especially in mid-sized trades on the LSE, and slightly increases private
information while also shifting private information processing to the LSE. On days with a
positive tone of public information, I find no significant change in liquidity, again a strong
increase in trading activity, and overall less private information impounded into markets
but a significant shift of the remaining private information from Chi-X to the LSE. Also
the total contribution to price discovery shifts to the LSE by 1.47%. One key finding
is that negative and positive public information have an asymmetric impact on trading.
For instance, Tetlock (2007) only finds a significant relation between pessimistic public
information and S&P 500 returns, also I find significant asymmetric reactions to good
and bad news in Chapter 2. Generally, it is expected that informed trading gravitate to
the most liquid market (Chowdhry and Nanda, 1991). Through the arrival of news, the
information environment and information processing of market participants change. The
increase in trading intensity hints at differential interpretation by market participants both
on positive and negative days. Positive information might be difficult to process such that
aggregate private information slightly falls while the impact of positive public information
is not strong enough to reduce competition for liquidity supply catering the increased need
for liquidity of liquidity demanders. I find that trades in the mid-size trade category (500
– 1,999 shares) have to pay a significantly higher effective spread, compared to the normal
difference, on Chi-X than on the LSE on positive days. Traders that are informed and have
the information processing capability for positive information then move to the LSE, the
more liquid market consistent with Chowdhry and Nanda (1991) and consistent with the
trade innovation results.
I find on both markets a reduction in liquidity on negative days. Competition for liquidity supply reduces, possibly as a result of liquidity suppliers safeguarding themselves
against better informed traders who are able to process the negative public information
correctly. Liquidity supply is not sufficient to cater the growth in liquidity demand. Also
more private information than normal is impounded into the market with again a significant shift from Chi-X to the LSE which is consistent with informed traders drawn to the

100

3 Fragmented Markets and Public Information

LSE (Chowdhry and Nanda, 1991). Robust measures are confirmed through a significant
increase in the simple price impact on the LSE. However, trading in FTSE 100 stocks is
overall still highly liquid on days with high levels of public information.

3.7 Conclusion
In this chapter, I study the effect of positive and negative firm specific public information
on trading in FTSE 100 constituents. The analysis comprises the LSE and Chi-X, the
two markets that account for the major part of non-OTC trading in FTSE 100 stocks.
Individual order books as well as characteristics of market fragmentation are examined in
this chapter. I find an asymmetric reaction to public information. Liquidity only decreases
for a negative tone of public information whereas trading activity increases strongly for
any type of public information. Price discovery shifts to the LSE on positive days and
more private information is impounded on the LSE on negative days. My findings are
consistent with existing literature (Kim and Verrecchia, 1991, 1994) within the individual
order books and also on the fragmentation characteristics (Chowdhry and Nanda, 1991).
Informed trading gravitates to the LSE, the most liquid market for FTSE 100 trading.
Overall, results also show that markets for FTSE 100 constituents are highly liquid and
stocks are actively traded based on relatively efficient price discovery processes. In practice, traders spend a considerable amount of money to subscribe to newswires of Thomson
Reuters, Bloomberg, or Dow Jones. Such newswires represent much of the real-time public information traders receive. I find that it is worthwile to observe the tone of public
information to be able to adjust trading and order routing decisions. The study in this
chapter confirms the important role that public information has in finding the efficient
price in equity markets and today’s computerized and fragmented trading landscape.
The perspective of the next chapter becomes broader and is not focused on market microstructure but analyzes how information production influences the comovement of international equity markets. It studies markets’ important information processing function
on an international level.

3.7 Conclusion

101

Table 3.12: Trading Activity and Public Information. Table 3.12 provides regression results for
trading activity measures. All measures are first aggregated on a daily basis per firm, then included
in the regression model. Panel A reports regression results on the number of trades (#Trades), the
natural logarithm of volume in kGBP (lnVolume), the natural logarithm of quantity in kShares
(lnQuantity), the average trade size in volume, the average trade size by shares traded, and market
shares by volume and quantity. Ratios for volume and quantity are calculated on raw values not
the natural logarithm of numbers. Panel B reports results on the number of trades per day and firm
splitted by trade size categories measured in the number of shares traded. The relevant variables are
dummy variables for positive and negative public information days. Table 3.12 presents results for
the LSE and Chi-X as well as the differences (Diff) and ratios (Ratio) between the LSE and Chi-X.
The regression model controls for tick size differences between the LSE and Chi-X, includes month
of the year dummy variables, and is estimated with firm-fixed effects. Spreads are measured in basis
points (bps). Robust t-statistics for the significance of the differences between the LSE and Chi-X
are reported in parantheses. ‘a’ indicates significance at the 1% level and ‘b’ indicates significance at
the 5% level.
Measure

Panel A: Trading Intensity and Trade Measures
LSE
pos

#Trades
Coeff.
t-stat

neg

Chi-X
pos

neg

Diff
pos

neg

Ratio
pos
neg

217.56a
(8.29)

256.54a
(7.14)

117.22a
(7.36)

105.62a
(4.55)

100.34a
(5.47)

150.92a
(6.23)

0.04a
(2.93)

0.01
(0.42)

lnVolume
Coeff.
t-stat

0.17a
(10.82)

0.13a
(9.54)

0.13a
(9.43)

0.11a
(8.11)

0.21a
(10.63)

0.15a
(9.05)

0.12a
(4.04)

0.06 b
(2.11)

lnQuantity
Coeff.
t-stat

0.15a
(10.25)

0.15a
(10.70)

0.11a
(8.60)

0.13a
(9.11)

0.19a
(10.09)

0.16a
(10.27)

0.12a
(4.06)

0.06 b
(2.13)

582.79a
(2.99)

301.37a
(3.46)

193.34 b
(2.45)

118.04a
(2.85)

389.46a
(3.13)

183.33a
(2.88)

0.03a
(3.51)

0.02 b
(2.42)

47.63
(1.53)

142.15a
(5.31)

13.38
(1.07)

56.05a
(4.09)

34.25
(1.66)

86.10a
(4.28)

0.03a
(3.52)

0.02 b
(2.46)

Market Share (Volume)
Coeff.
t-stat

0.87%a
(6.23)

0.30% b
(2.38)

-0.87%a
(-6.23)

-0.30% b
(-2.38)

1.73%a
(6.23)

0.60% b
(2.38)

Market Share (Quantity)
Coeff.
t-stat

0.87%a
(6.24)

0.30% b
(2.39)

-0.87%a
(-6.24)

-0.30% b
(-2.39)

1.73%a
(6.24)

0.60% b
(2.39)

Trade Size (Volume)
Coeff.
t-stat
Trade Size (Quantity)
Coeff.
t-stat

continued on next page . . .

3 Fragmented Markets and Public Information

102

. . . continued from Table 3.12
Trade Size Category

Panel B: Number of Trades
LSE
pos

< 500
Coeff.
t-stat
500 − 1, 999
Coeff.
t-stat
2, 000 − 4, 999
Coeff.
t-stat
5, 000 − 9, 999
Coeff.
t-stat
≥ 10, 000
Coeff.
t-stat

neg

Chi-X
pos
neg

Diff
pos

neg

Ratio
pos
neg

66.37a
(6.74)

45.12a
(4.44)

51.29a
(5.21)

16.48
(1.15)

15.08 b
(2.49)

28.64 b
(2.37)

0.01
(0.88)

-0.00
(-0.09)

79.53a
(9.28)

84.80a
(8.07)

41.37a
(7.85)

43.16a
(6.84)

38.16a
(5.48)

41.64a
(5.97)

-0.00
(-0.06)

-0.06
(-1.13)

43.04a
(7.55)

58.96a
(6.92)

15.63a
(4.95)

24.31a
(5.05)

27.42a
(6.28)

34.65a
(6.34)

0.37a
(2.71)

0.11
(1.07)

19.59a
(3.93)

33.60a
(4.53)

6.44 b
(2.55)

11.94a
(3.37)

13.14a
(3.64)

21.66a
(4.15)

1.08a
(4.09)

0.70a
(3.43)

9.03
(1.46)

34.06a
(3.09)

2.49
(1.74)

9.73a
(2.95)

6.54
(1.26)

24.33a
(2.75)

0.87a
(3.00)

0.46
(1.61)

3.7 Conclusion

103

Table 3.13: Price Discovery and Public Information. Table 3.13 provides regression results (cf.
Equation 3.3) for information measures based on Hasbrouck (1991a,b, 1995). All measures are
computed on a daily basis per firm and then included in the regression model. Results on the LSE,
Chi-X, and the differences between the two markets (Diff) are presented. Panel A reports results
on trade based price discovery: the share of trade based price discovery in total price discovery
(% Trade Based) and the permanent impact of trade innovation in basis points (Trade Innovation).
Panel B reports descriptives on quote based price discovery: the overall share of quote based price
discovery in total price discovery and the share of the LSE and Chi-X respectively in quote based
price discovery. Panel C provides information about the total contribution to price discovery of the
LSE and Chi-X. Robust t-statistics for the significance of the differences between the LSE and ChiX are reported in parantheses. ‘a’ indicates significance at the 1% level and ‘b’ indicates significance
at the 5% level.
Panel A: Trade Based Price Discovery
% Trade Based

Trade Innovation

Positive

Negative

Positive

Negative

LSE
t-stat

0.05%
(0.20)

0.02%
(0.12)

-0.06
(-1.92)

0.06 b
(2.47)

Chi-X
t-stat

0.02%
(0.11)

0.10%
(0.78)

-0.19a
(-2.74)

-0.04
(-1.09)

Diff
t-stat

0.03%
(0.13)

-0.09%
(-0.36)

0.13 b
(2.48)

0.10a
(2.61)

Panel B: Quote Based Price Discovery
% Quote Based

Positive

Negative

LSE
t-stat

0.85%a
(2.59)

0.44%
(1.18)

Chi-X
t-stat

-0.85%a
(-2.59)

-0.44%
(-1.18)

Diff
t-stat

1.17%a
(2.59)

0.89%
(1.18)

Overall
t-stat

Positive

Negative

-0.07%
(-0.20)

-0.16%
(-0.48)

Information Shares

Panel C: Total Contribution to Price Discovery
Fraction of PD
Positive

Negative

LSE
t-stat

0.73%a
(3.09)

0.25%
(0.97)

Chi-X
t-stat

-0.73%a
(-3.09)

-0.25%
(-0.97)

Diff
t-stat

1.47%a
(3.09)

0.52%
(0.97)

Chapter 4
Comovement in International Equity
Markets and Public Information
4.1 Introduction
Over the last decades financial markets have become more globalized than ever. Financial
instruments are traded 24/7 around the globe on market places based in small emerging
economies as well as large developed countries. And still, we lack knowledge about price
formation and how information influences financial markets specifically in different countries. Particularly in a globalized world with linked financial markets, it is essential to
understand how markets process information in order to foster efficient and stable financial markets for the future. One aspect of interest is how firm specific stock returns vary in
relation to market and industry returns. Literature suggests that a higher portion of firm
specific stock price variability in the overall stock price variability is associated with more
efficient stock markets (Roll, 1988; Durnev et al., 2003).
A firm’s variability in stock prices can be either idiosyncratic, industry specific, or market driven. Idiosyncratic variability is often also called firm specific variability or firm
specific volatility. The market driven component is also called stock market or systematic variability or volatility. The different components of stock price variability relate to
eacher other. Market synchronicity, stock return comovement, or only comovement, is a
measure that can be calculated in different ways but fundamentally expresses the degree to
which stocks of individual firms in one market move together. It is the fraction of stock
price changes that are explained by market and industry changes. Some studies also investigate how entire countries’ stock markets move together. However, I base this chapter on

106

4 Comovement in International Equity Markets and Public Information

the former definition of comovement. A change in comovement is a change in the relation
of firm specific volatility to industry and market wide volatility. It is still not entirely clear
what some underlying drivers of comovement are. Recent research suggests that information production and a country’s information environment, e.g. transparency and other
institutional settings, significantly influence stock return comovement (Brockman et al.,
2010).
In this chapter, I am first interested whether changes in the overall firm specific public
information flow are reflected in the synchronous movement of entire markets. Second,
I study how country specific financial development and transparency characteristics influence the association of the firm specific information flow and stock return comovement.
In contrast to existing research, the variability of firm specific information production is
directly measured. Generally, one major source for firm specific public information flow
is the Thomson Reuters newswire service. As in the previous chapters, it is again a suitable
source for firm specific public information since “market participants use this news service
on a regular basis, along with Dow Jones News Service and perhaps a few other newswires,
as a prime news source for economic decision making” (Berry and Howe, 1994).
Generally, I find that the overall firm specific information flow has a significant influence on stock return comovement. An increase in relative firm specific public information
reduces stock return comovement, thus increases idiosyncratic stock price variability. In
addition, the strength of this association significantly depends on a country’s institutional
setting. The quality of a firm’s information environment and legal protection of outside
investors significantly determine how strong the association between firm specific information flow and stock return comovement is. Transparency and effective investor protection
reduce the relation between the firm specific public information flow and stock market
synchronicity.
The remainder of the chapter is organized as follows. Section 4.2 introduces related
work. Section 4.3 provides a detailed description of the newswire data set, stock market data, additional cross sectional per firm and country data, and the sample selection
process. Section 4.4 presents stock return comovement and news comovement measures
while Section 4.5 introduces the regression framework and provides results. Section 4.6
finally concludes the chapter.

4.2 Related Work

107

4.2 Related Work
One central paper for this chapter is the work by Campbell et al. (2001) who develop the
comovement measure which is used in this chapter. In their paper, they show that firm specific volatility has increased over the last decades in the US market. Their data ranges from
1962 to 1997, a long time series to analyze trends. When firm specific volatility is high,
stock market comovement is low and vice versa. Intuitively, high average firm specific
volatility corresponds to high average firm specific risks which is how some papers, that
base on the Campbell et al. (2001) measure, interpret firm specific volatility. With respect
to the research question, many papers that use the Campbell et al. (2001) comovement
measure focus on either the time series or cross sectional characteristics of comovement.
The source of time trends in stock return comovement has been of interest to numerous
research papers. Cao et al. (2008) compute one measure of idiosyncratic firm volatility,
comparable to the comovement measure used in this chapter, with high idiosyncratic firm
volatility being equivalent to low comovement. In contrast to my analysis, they base their
calculation on the market and firm volatility leaving out the industry component. The past
decades have shown a time trend increase in idiosyncratic firm volatility in the US stock
market which they explain with an increase in growth options or growth opportunities
of firms. In their regression framework, time trends become insignificant once proxies
for growth opportunities are included in the models. In an analysis of the G7 countries –
Canada, France, Germany, Italy, Japan, the United Kingdom, and the United States – Guo
and Savickas (2008) make essentially the same argument, that idiosyncratic firm volatility is
related to changes in investment opportunities which are driven by growth opportunities.
They also find a high correlation of average per market measured idiosyncratic volatility
among the G7 countries which the other way around also means a high correlation of
stock market comovement. In another study of the US market, Fink et al. (2006) find that
idiosyncratic risk computed with the Campbell et al. (2001) comovement measure is driven
by firm age at initial public offerings (IPO) plus the increase in the number of IPOs over
time. During the last decades, the average age of firms at IPOs has decreased dramatically.
Since younger firms tend to be riskier, this effect in combination with a higher number of
IPOs has driven the time trend in comovement.
Consistently, Brown and Kapadia (2007) study the time trend of comovement in the
United States based on a very long time series from 1963 to 2004. They use the relation
of firm specific risk to market wide risk as a type of comovement measure. The authors

108

4 Comovement in International Equity Markets and Public Information

interpret low stock return comovement as the presence of higher firm specific risk in relation to the market wide risk. One driver of time trends is, according to their analysis,
the change in characteristics of publicly traded firms. Over time smaller companies and
riskier industries have been listed on the public stock market increasing firm specific risk
and thus reducing the overall stock market synchronicity. They do not use the Campbell
et al. (2001) measure of comovement but directly compare their results to the results of
Campbell et al. (2001) which are consistent. Irvine and Pontiff (2009) explain the increase
in firm specific volatility over the last decades specifically in the US market with an increase in economy wide competition. In the light of the Brown and Kapadia (2007) results,
Irvine and Pontiff (2009) conclude that “financial innovation allows small, risky firms to
raise capital, thus inducing greater economy-wide competition”. Irvine and Pontiff (2009)
extend their analysis to international markets and find the same effect. In non-US markets,
economy-wide competition also increases firm specific volatility thus reduces stock market
comovement.
In another study of the US stock market, Chun et al. (2004) propose that an increase
in firm specific volatility, thus a reduction in comovement, is related to the dramatic development of information technology and its increased use in firms. They argue that, like
electricity a hundred years ago, information technology has become a general purpose
technology. Information technology improves production processes and puts more importance to intangible outputs. Firm specific, or idiosyncratic, volatility increases since
possibilities for improvement based on information technology are used differently by
firms and result in higher heterogeneity of firm performance. Chun et al. (2004) also find
that industries which rely stronger on information technology exhibit higher firm specific
volatility than other industries.
Hamao et al. (2003) study firm specific and market wide risk in the Japanese stock market, one of the few analyses that does not include the US market. They also use the Campbell et al. (2001) measure. In contrast to the United States, the Japanese stock market shows
a strong decrease in firm specific volatility after its crash in 1990. Hamao et al. (2003) attribute this decrease in firm specific volatility, or increase in comovement, to homogeneity
in the performance of Japanese firms and the protection from bankruptcies which resulted
in a “lack of creative destruction [...] and added to the difficulty of sorting out healthy
firms in the capital allocation process” (Hamao et al., 2003). Thus, Hamao et al. (2003)
show that the time trend properties of comovement do not need to necessarily be the same
in different countries.

4.2 Related Work

109

One stream of literature links stock return comovement with firm specific information
using information related proxies. In the interpretation of this literature, the focus often lies on the firm level, i.e. specific firm attributes and not the average firm. Durnev
et al. (2003) analyze the US market over the the years 1983 to 1995. They study whether
price informativeness, proxied through accounting measures such as future earnings, relates to firm specific stock price variation which in their analysis is the complement to
comovement. Their definition of firm specific stock price variation is close to mine of
“firm-specific price variation as the portion of a firm’s stock return variation unexplained
by market and industry returns” (Durnev et al., 2003). One major result is that higher firm
specific variation, lower comovement, indicates more informative stock market prices.
“Firm-specific variation in U.S. stock returns most likely reflects the capitalization of firmspecific information into stock prices” (Durnev et al., 2003). Durnev et al. (2004) analyze
the relation of the efficiency of corporate investment1 , or broadly speaking efficient capital
allocation, with firm specific return variation in the US market from 1993 to 1997. Higher
economic efficiency of corporate investment positively correlates with firm specific stock
return variation. According to Piotroski and Roulstone (2004), an interpretation of the
Durnev et al. (2004) results is that “a stronger flow of firm-specific information should
allow for greater monitoring and reduced information asymmetry between insiders and
outsiders, the observed relations between low synchronicity and efficient capital allocation decisions indirectly support the interpretation that synchronicity reflects the flow of
firm-specific information”.
Piotroski and Roulstone (2004) present another analysis of the US market. The focus
on one country ensures that the results are not driven by country differences. Their study
reveals that analyst forcast activity increases comovement. Analysts are by design outsiders with relatively little firm specific information in contrast to insiders. They increase
industry level information which reduces firm specific volatility. Insider trading, insiders having presumably firm specific information, on the other hand reduces stock return
comovement. Their results suggest that firm specific information reduces comovement
while comovement is increased by market wide or industry wide information. Hameed
et al. (2010) present an analysis of the US market from 1984 to 2007. If one stock is heavily covered by analysts, other less covered stocks within the same industry follow. They
“document that the stock returns of firms followed by many analysts contribute to the syn1

Their measure for the efficiency of corporate investiment is Tobin’s marginal q ratio.

110

4 Comovement in International Equity Markets and Public Information

chronicity of stock returns” (Hameed et al., 2010), not necessarly for the whole market but
for firms whose fundamentals are close. Interestingly, this behavior is more pronounced
if the base level of analyst coverage is low. However, Hameed et al. (2010) restrict their
finding to the US market and speculate that market behavior could be different in other
markets, especially in emerging economies. Their study is consistent with Piotroski and
Roulstone (2004) who postulate that analysts disseminate mainly industry information.
Hutton et al. (2009) provide evidence that opaqueness is linked to stock return comovement using US market data from 1991 to 2005. “When less firm specific information is
publicly available, fewer observable reasons exist for individual stock returns to depart
from broad market indexes and market synchronicity increases” (Hutton et al., 2009). In
contrast to other papers they study opaqueness and comovement directly at the firm level
using earnings management as a firm specific opaqueness measure. “Firms with opaque
financial reports have stock returns that are more synchronous with the market” (Hutton
et al., 2009).
All studies above mostly study the differences of firms within the US market, including the linkage between information and comovement, or the time trend proporties of comovement. I focus on a determinant of comovement that is neither a time trend nor purely
country specific but driven by the time-varying characteristics of information production,
in particular of firm specific information. The association of time-varying variables might
differ as a result of cross sectional country characteristics, e.g. institutional settings. Consequently, the area of literature that studies different characteristics among countries, their
relation to comovement, and the association with a firm’s information environment is also
important for this chapter.
Jin and Myers (2006) provide a study of 40 countries around the world analyzing the
years 1990 - 2001. They find that opaqueness in a country increases comovement, called
R2 in their paper. Weaker control rights and lower availability of firm specific information
shift some firm specific risk from investors to managers. If firms are less transparent, insiders, for instance managers, can more easily divert cash flows to themselves. However,
in doing so they also carry more firm specific risk since they can divert more if inside firm
specific information is positive and less when it is negative. Insiders carrying more firm
specific risk then increases the synchronicity of stock returns. In short, Jin and Myers
(2006) provide evidence that the information environment and a firm’s intrinsic transparency level to outside investors can have a significant influence on the comovement of
a country’s stock market. One conclusion which can be drawn from their findings is that

4.2 Related Work

111

a high amount of firm specific information might result in a reduction of stock return
comovement.
Morck et al. (2000) study the impact of institutional development on the degree of comovement. Their sample includes poor emerging countries and rich developed economies.
A particular institutional feature of interest in their work is the strength of property rights.
The lack of strong property rights of outside investors in poor countries explains high
stock market comovement. Morck et al. (2000) conjecture that those effects can be attributed to less informed trading on proprietary firm specific information. If the political
class has more influence on stock prices through direct influence on firms, the uncertainty
of future returns increases. In addition, informed outside traders might not even be able
to keep their profits based on a lack of property rights. Both factors discourage informed
trading. Their study shows that a country’s institutional setting might affect how much
private firm specific information is capitalized into stock prices, also among developed
countries. Li et al. (2004) study the relation between comovement and financial market
openness for different emerging markets. They find that lower comovement is associated
with higher financial market openness. Using emerging markets provides the opportunity
for an acadamic analysis to still find markets that are not financially open in contrast to
developing countries. One of their proxies for financial market openness is “good government” which subsumes the rule of law, efficiency of the legal system, and freedom from
corruption. Karolyi et al. (2009) study stock markets of 40 countries around the world
from 1995 to 2004 including both developed and emerging economies. Consistent with
existing literature, they find that comovement is larger in countries with weak investor
protection and opaque information environments. They not only investigate stock return
comovement but also liquidity comovement and find a strong positive correlation between
both.
Bushman et al. (2004) analyze the determinants of corporate transparency which they
define as “the availability of firm-specific information to those outside publicly traded
firms”. They find two main factors that characterize a countries firm specific information
environment, financial transparency and governance transparency. The financial transparency factor, for instance, captures information dissemination by media outlets. The
governance factor is strongly related to a country’s legal system with higher governance
transparency in common law countries. High financial transparency is driven by low state
ownership of firms and low risk of state expropriation. One additional finding is “that financial transparency is significantly higher where firms are larger” (Bushman et al., 2004).

112

4 Comovement in International Equity Markets and Public Information

Their paper shows that political and legal characteristics of a country substantially influence a firms information environment.
Wurgler (2000) studies characteristics of efficient capital allocation across 65 countries
based on approximately 30 years of return observations for each individual country. He
finds a negative correlation between efficient capital allocation and stock return comovement which is interpreted as a measure of how much firm specific information is incorporated into stock prices. In addition, one result is that efficient capital allocation correlates
positively with the legal protection of minority investors, a financial development indicator. Bris et al. (2007) provide additional evidence that institutional characteristics of
countries influence the firm specific variation of stock returns, thus also comovement, and
as such the informativeness of stock prices. Their analysis comprises of 46 stock markets
from all over the world over the years 1990 to 2001. Bris et al. (2007) find that less negative firm specific information measured by lower idiosyncratic stock return variability is
incorporated into stock prices when short selling is restricted.
In the paper that is probably the closest to this chapter, Brockman et al. (2010) hypothesize that the comovement in stock returns is driven by time-varying information
production. Based on recent research (Veldkamp, 2005) which connects information production and aggregate economic activity, Brockman et al. (2010) connect stock return comovement with measures of aggregate economic activity, for instance with gross domestic
product (GDP) growth. Veldkamp (2005) presents a theoretical model which predicts that
information production is high during times of economic expansion and that information
production is low during times of economic decline. Since demand is lower for information during times of economic decline, costs for information rise as the fixed costs of
information have to be apportioned to a lower number of information demanders. As a result “with less firm-specific information available comovement increases” (Brockman et al.,
2010). Brockman et al. (2010) find in their analysis a negative relation of economic growth
and stock return comovement. Using economic growth as a proxy for information production, its time-varying characteristics have an influence on stock return comovement.
A low amount of firm specific news drives an increase in stock return comovement. To
measure comovement driven by aggregate economic activity as a proxy for information
production, Brockman et al. (2010) “study the relation between economic activity and
comovement while jointly controlling for country and time effects using panel data”.
A theoretical model that also motivates the analysis in this chapter is presented by Veldkamp (2006) who analyzes the market for information and its relation to comovement.

4.3 Data and Sample Selection

113

Her model predicts that comovement increases when many investors only demand a subset of information as a result of costly information. However, when the number of information signals increases and information is additionally available for more stocks then
comovement decreases. Less investors infer information about an asset from another asset’s information. High levels of information production should associate with low levels
of comovement which is important when countries exhibit different information environments, e.g. influenced through institutional settings. Also, a relatively low amount of firm
specific information should increase stock return comovement.
Based on previous research, especially Brockman et al. (2010), I first hypothesize that
an increase in the directly measured relative per firm information production in a market, as one factor, reduces stock return comovement after controlling for country effects
and time trends. Second, this relation should vary with financial development and transparency characteristics of individual countries. I remove time and country specific effects
comparable to Brockman et al. (2010) through time trend and per country controls in
the regression models and, in contrast to existing research, I apply a direct proxy for firm
specific information based on Thomson Reuters newswire messages.

4.3 Data and Sample Selection
The data in this chapter bases on manifold sources. The major data source is firm specific
stock market and news data which are both cross-sectional as well as time-series data. In
addition, pure cross-sectional per country data as well as cross-sectional per firm data are
included in the analyses.

4.3.1 Market Data
I retrieve daily per firm prices and volumes as well as foreign exchange data for the years
2005 to 2009 from the Thomson Reuters DataScope Tick History archive through SIRCA
as in the previous chapters. Sample stock market data can be found in Appendix B. Per
firm data includes stock split information and dividend payments. All prices are reported
in local currencies in the raw data. Daily returns are simple returns and calculated stock
split and dividend adjusted. Trading volume is derived from local currency trading volume
in combination with the daily US dollar foreign exchange rate and reported in US dollars.
Daily excess returns for firms, industries, and countries are calculated in excess of daily

114

4 Comovement in International Equity Markets and Public Information

one-month US Treasury Bill returns derived from Kenneth French’s data library2 . To
control for extreme returns or data recording errors, excess returns are winsorized at 99%
and 1% comparable to Brockman et al. (2010).

4.3.2 News Data
This chapter’s analysis is based on RNSE news data as presented in Chapter 2 Section 2.4,
the data which are used throughout this thesis. The entire news data for firms traded
worldwide are used as a basis for this chapter. It is again important to recall that one news
item is scored separately for different firms. A more detailed description of RNSE data
fields is available in Appendix C. This chapter bases on the sentiment and relevance scores
of RNSE data. Again, sentiment reflects the stock specific tone of one news item and is
either positive (1), negative (-1), or neutral (0). Relevance is a stock specific score between 0
and 1 (including 1 and not 0) for a single message. The closer relevance is to one, the more
relevant a news message is for a particular firm.
News data are aggregated to daily per firm measures for further analyses in a two step approach. First, I weight sentiment with the relevance of a news message. I compute for each
single message the product of sentiment and relevance. Second, I calculate the average daily
value of this product (‘sentrel’) for a specific firm. If no news message arrives for a firm
on a specific day, 0 is assigned as a value to this specific firm and day combination. Daily
aggregate measures range between -1 and 1. To check for robustness of the analyses, daily
aggregated values are also calculated without weighting sentiment measures by relevance.
If news variables are zero, news volatility is low, if news variables indicate firm specific
news, news volatility is driven up. If no news arrives, news variability is by definition
zero, in line with intuition. The analysis is based on 4,442,097 raw news messages items in
the final sample of firms and countries over the years 2005 to 2009 (cf. Section 4.3.4).

4.3.3 Cross-Sectional Data
The study in this chapter also uses country and firm specific cross-sectional data based on
literature presented in Section 4.2. Country specific variables consist of the Corruption
Perceptions Index (CPI), the ICT Development Index, stock market size, per capita gross
domestic product (GDP), an index of antidirector rights, an index of accounting quality,
and finally whether a country is a civil law country or not.
2

http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html.

4.3 Data and Sample Selection

115

For the CPI, the 2007 ranking is used since year to year changes in the CPI are not
comparable. In addition, 2007 lies in the middle of the sample period. The CPI measures
perceived corruption on a scale from 0 to 10 and is retrieved from Transparency International3 , a non-governmental organisation fighting against and reporting on corruption
globally. A value of 10 would characterize a country without any corruption. The ICT
development index is compiled by the International Telecommunication Union (ITU), a
United Nations agency. The index measures the advancement of information and communication technology in a country. I derive the 2007 numbers which are the most current
data in the 2009 ‘Measuring the Information Society’ report (International Telecommunication Union, 2009). Stock market size is retrieved from the World Federation of Exchanges (WFE)4 and measured by the average entire domestic market capitalization in US
dollars over the years 2005 to 2009 for each country in the sample.
The data for antidirector rights and accounting quality are derived from Andrei
Shleifer’s data sets.5 “Antidirector rights measure how strongly the legal system favors minority shareholders against managers or dominant shareholders in the corporate decisionmaking process, including the voting process” (La Porta et al., 1998). Antidirector rights
consist of six rights, if each is granted to investors in a country, the index variable is six,
if investors have not a single of the six rights, the variable takes zero, and otherwise the
number of given rights is counted. Thus, the antidirector rights index ranges from zero
to six with higher values being better. The first right is the right for absent voting, for
instance via mail, for an investor to be able to execute voting rights. The second area is
whether shareholders need to deposit shares around shareholders’ meetings in order to
execute voting righs, if they do, minority shareholders are discouraged to come to shareholders’ meetings and vote. The third is the right for cumulative voting or proportional
board representation which protects minority shareholders. The fourth is whether legal
mechanisms against directors exist for minority shareholders. The fifth is whether shareholders’ have a preemption of new issues. And the final antidirector right is the necessary
percentage of share capital to call an extraordinary shareholders’ meeting. Since a percentage cannot be directly expressed in terms of whether a right is given or not, La Porta et al.
(1998) introduce a barrier of 10%. Below and at it is counted as one for the total calculation
of the antidirector measure and above it is calculated as zero.
3

http://www.transparency.org/.
http://www.world-exchanges.org/statistics/.
5
http://www.economics.harvard.edu/faculty/shleifer/dataset/.

4

116

4 Comovement in International Equity Markets and Public Information

The accounting quality index is constructed by La Porta et al. (1998) and described as “an
index created by examining and rating companies’ 1990 annual reports on their inclusion
or omission of 90 items falling in the categories of general information, income statements,
balance sheets, funds flow statement, accounting standards, stock data, and special items”
(La Porta et al., 2000). If the numerical value of the accounting quality index is higher, the
respective country has a higher accounting quality. Per capita GDP is derived in US dollars
from the Worldbank database.6 Whether a country is a civil or common law country is
compiled from own research.7
In addition to cross-sectional country data, I retrieve the Thomson Reuters Business
Classification8 (TRBC), a market oriented schema to globally classify firms. TRBC includes four hierarchies: 10 economic sectors, 25 business sectors, 52 industry groups, and
124 industries. For the purpose of this analysis, I rely on the TRBC business sector hierarchy level to differentiate firms by their industry affiliation. A descriptive summary of the
classification for this chapters’s sample (see Section 4.3.4) can be found in Table 4.1. Most
stocks are traded in the ‘Banking and Investment Services’ business sector with an average
yearly trading volume of almost four trillion US dollars. In the ‘Energy’ sector 322 stocks
are traded, 200 less than in ‘Banking and Investment Services’. However, the average yearly
trading volume is only 500 billion US dollars lower than in the ‘Banking and Investment
Services’ category which increases the average per firm trading volume considerably. The
highest average yearly per firm trading volume can be found in the ‘Telecommunication
Services’ sector with a little more than 12 billion US dollars. The by far smallest average
yearly per firm turnover is found in ‘Investment Trusts’ with 0.7 billion US dollar which
consequently has the effect that those firms have barely an influence on the comovement
measures. The next smallest turnover measure is observed for ‘Industrial Services’ averaging 2.53 billion US dollars per year and firm.

6

http://databank.worldbank.org/.
Information about a country’s legal system can be, for instance, found in the CIA World Factbook
database, https://www.cia.gov/library/publications/the-world-factbook/index.html.
8
http://thomsonreuters.com/products_services/financial/thomson_reuters_indices/trbc/.
7

Energy
Chemicals
Mineral Resources
Applied Resources
Industrial Goods
Industrial Services
Industrial Conglomerates
Transportation
Automobiles and Auto Parts
Cyclical Consumer Products
Cyclical Consumer Services
Retailers
Food and Beverages
Personal and Household Products and Services
Food and Drug Retailing
Banking and Investment Services
Insurance
Real Estate
Investment Trusts
Healthcare Services
Pharmaceuticals and Medical Research
Technology Equipment
Software and IT Services
Telecommunication Services
Utilities

Energy
Basic Materials

Telecommunication Services
Utilities

Technology

Healthcare

Financials

Non-Cyclical Consumer Goods and Services

Cyclical Consumer Goods and Services

Industrials

Business Sector

Economic Sector
322
102
212
57
283
236
25
130
64
184
268
163
139
69
39
522
159
174
28
231
239
279
231
104
148

Number of Stocks

3,572.64
651.72
1,759.84
163.26
1,205.56
597.97
290.75
546.31
518.41
728.21
1,042.28
1,046.78
1,016.39
313.11
391.30
3,994.36
1,464.81
583.96
19.71
631.44
1,495.20
2,139.67
1,568.37
1,276.24
993.54

Avg. Trading Volume (bnUSD)

Table 4.1: Descriptive Statistics Thomson Reuters Business Classification. This table presents the Thomson Reuters Business Classification
Scheme used to define industries for the comovement calculation. ‘Number of Stocks’ reports the overall number of stocks in the entire sample
that belong to one business sector. The average yearly trading volume is reported in billion US dollars.

4.3 Data and Sample Selection
117

118

4 Comovement in International Equity Markets and Public Information

4.3.4 Sample Selection
The selection of securities in this study is based on available Thomson Reuters newswire
messages, only stocks and countries that are covered by RNSE data from 2005 to 2009 are
taken into consideration for this analysis. Stocks in the final sample have to have at least
one news message per year such that I know that a firm is still covered by RNSE archive
data. A country requires at least 10 traded firms to stay in the sample while existence of
stock split information and dividend data is another necessary condition. After considering all conditions, the sample consists of 4, 408 securities traded in 23 countries over the
years 2005 to 2009 with 4,442,097 raw news messages. Decriptive market statistics for the
sample are provided in Table 4.2 and statistics for the Thomson Reuters Business Classification of the sample can be found in Table 4.1 which shows that the sample also covers a
broad section of industries. The sample comprises with Argentina, India, and Indonesia
three developing countries based on the Worldbank classification as of August 2010.9
Market summary statistics in Table 4.2 show that the firm sample is by far the largest
in the United States comprising of approximately three quarters of all sample stocks. This
phenomenon is also evidence for the fact that most firm specific public information flow,
here proxied through Thomson Reuters news messages, is disseminated for US listed firms
in contrast to the rest of the world. The average yearly per firm trading volume is with a
bit less than five billion US dollars quite small for the United States compared to the rest
of financially developed countries. It seems as if a huge number of stocks, that are relatively small in comparison to the average firm size in the market, is followed by Thomson
Reuters news in the United States which does not appear to be the case in other countries.
The smallest per firm yearly trading volume is found in the Argentinian sample with only
211 million US dollars. Argentina is also one of the three countries in the sample that are
classified as a developing country by Worldbank standards. The global overall volume in
share trading on regulated exchanges is, according to the World Federation of Exchanges
(WFE), on average 83.3 trillion US dollars per year from 2005 to 2009. All firms in my
sample have a combined average yearly trading volume of 28.1 trillion US dollars which is
almost 34% of global equity trading based on 23 countries with 4,408 traded firms that are
the basis for my analysis.
Domestic market capitalization of the entire market is not based on sample firms but
reports values for a country’s entire regulated stock market as reported by the World Fed9

http://data.worldbank.org/about/country-classifications/.

4.3 Data and Sample Selection

119

eration of Exchanges. The average domestic market capitalization of all countries in the
sample is approximately 35.6 trillion US dollars. The overall market capitalization of exchanges that report to the World Federation of Exchanges is on average 46.6 trillion US
dollars over the years 2005 to 2009. The 23 sample countries constitute almost 80% of the
world’s entire market capitalization and as such those countries should be representative
for global investor behavior.
The yearly average annual returns, dividend and stock split adjusted, compare well to
Brockman et al. (2010) considering that the sample periods are different. The only strongly
negative average annual return is recorded for Ireland at a value of -7.70%. The major driver
for that development is the financial crisis which had a dramatic negative effect on returns
in 2008 and 2009. The annual returns for 2008 and 2009 on average decrease the average
annual returns for most countries in the analysis. The highest average annual return rate is
observed for India at 57.44%, closely followed by Indonesia at 51.85%. Both countries are
developing countries as defined by the Worldbank and their stock markets should include
additional risks for outside investors in comparison to developed financial markets. The
equal weighted average over all countries is 10.50%, a reasonable number for the time
period from 2005 to 2009 without risk adjustments.

17,594
202,761
22,077
167,311
15,988
156,014
47,801
55,650
49,646
89,581
30,001
23,998
100,693
31,671
45,342
22,493
15,008
33,578
44,247
23,141
42,238
346,682
3,765,037

15
163
18
134
13
124
38
45
41
73
25
20
80
25
37
18
12
27
35
19
34
278
3,134
4,408

Argentina
Australia
Austria
Canada
Denmark
France
Germany
Greece
Hongkong (China)
India
Indonesia
Ireland
Italy
Netherlands
New Zealand
Norway
Portugal
Singapore
Spain
Sweden
Switzerland
United Kingdom
United States

Overall

5,348,552

Number of
Observations

Country

Number of
Stocks

6,373

211
4,654
3,521
5,902
4,748
13,388
24,228
1,726
16,712
1,402
1,417
2,955
13,584
25,899
453
13,490
4,388
4,661
26,000
16,301
31,522
12,849
4,735

Average Yearly Per Firm
Trading Volume (Mio. USD)

28,092,161

3,172
748,625
63,374
790,932
61,725
1,660,066
920,663
77,650
685,199
102,352
35,429
59,101
1,086,688
647,483
16,754
242,823
52,660
125,857
910,012
309,727
1,071,744
3,572,035
14,838,089

Yearly Trading
Volume (Mio. USD)

35,578,688

48,299
1,028,794
150,449
1,615,941
161,063
1,726,260
1,473,370
164,234
1,811,656
1,984,010
149,142
106,408
815,010
978,219
38,522
239,471
81,299
385,405
1,289,370
465,941
1,072,765
3,073,759
16,719,301

Domestic Market Cap
Entire Market (Mio. USD)

10.50

31.16
15.23
7.94
16.99
19.86
3.40
10.73
8.49
27.19
57.44
51.85
-7.70
2.24
5.30
-0.38
21.68
7.35
21.19
9.33
16.31
3.92
8.37
11.08

Average Annual
Returns (%)

Table 4.2: Market Summary Statistics. This table presents market statistics for the sample for the observation period from 2005 to 2009. The
domestic market capitalization is based on World Federation of Exchanges’ data for the entire domestic equity market. All other measures
are based on the firms in the sample over the years 2005 to 2009. Average annual returns are simple dividend adjusted returns for a country
compiled from trading volume weighted firm returns for the country return computation. Averages in the ‘Overall’ line are equal weighted
over all countries. All monetary measures are reported in US dollars.

120
4 Comovement in International Equity Markets and Public Information

4.4 Measures

121

4.4 Measures
4.4.1 Stock Market Comovement
To calculate stock market comovement, I resort to the definition of Brockman et al. (2010)
which is based on a decomposition into firm specific, industry specific , and market wide
volatility (Campbell et al., 2001) without the use of firm specific betas. Numerous other
research articles also use this volatility decomposition.10 Comovement measures are calculated separately for each country and each month. Since GDP growth data is only available by quarters for most countries, Brockman et al. (2010) compute their comovement
measure on a quarterly basis. I do not have such data restrictions and perform the decomposition analogous to Campbell et al. (2001) on a monthly level.
The first step of the Campbell et al. (2001) volatility decomposition is to calculate daily
weighted market excess returns and daily weighted industry excess returns. Excess returns
are calculated against the risk free rate represented by daily returns of one-month US Treasury Bills. Let c denote a market (country), i an industry, and j an individual firm. Days
are identified by the variable s. Let wi,c,s be the weight of an industry i in country c on
day s . The weight of firm j in industry i in a country c on day s is w j ,i,c,s . Let R j ,i,c,s
denote an individual firm’s excess return. Then, industry excess returns are defined as
Ri ,c,s =

X

w j ,i,c,s R j ,i,c,s .

(4.1)

j ∈i

Market, and as such country, excess returns are defined as
Rc,s =

X

wi,c,s Ri,c,s .

(4.2)

i∈c

In this study, returns are weighted by daily trading volume. The originally proposed
volatility decomposition is based on market value weights but the decomposition is valid
for any weighting scheme (Campbell et al., 2001). In addition, there should be not much
difference between weighting by trading volume and market value. Based on above returns, the three volatility components – market wide, industry specific, and firm specific
– are estimated monthly for each country. Let µc be the mean weighted market excess
10

Cao et al. (2008), Fink et al. (2006), and Irvine and Pontiff (2009) study die US market Hamao et al. (2003)
analyze the Japanese market and Guo and Savickas (2008) present a study on the G7 countries, Canada,
Germany, France, Italy, Japan, the UK, and the US.

122

4 Comovement in International Equity Markets and Public Information

return for country c over the entire sample period and days are denoted by s then MKTc,t ,
the market wide volatility of country c in month t , is computed as
MKTc,t =

X€

Rc,s − µc

Š2

.

(4.3)

s∈t

Industry volatility INDc,t of country c in month t is the weighted average industry volatility in month t and country c and defined as
INDc,t =

X

wi,c,s

X€

Ri,c,s − Rc,s

Š2

!
.

(4.4)

s∈t

i ∈c

Firm volatility FIRMc,t of country c in month t is the weighted average firm volatility in
month t and country c and defined as

FIRMc,t


!
X
X
X€
Š2 
R j ,i,c,s − Ri ,c,s
=
w j ,i,c,s
wi,c,s
.
i∈c

j ∈i

(4.5)

s∈t

Using above equations, comovement in the spirit of Brockman et al. (2010) for a country
c in month t is calculated as
COMVc,t = 1 −

FIRMc,t
MKTc,t + INDc,t + FIRMc,t

.

(4.6)

The comovement measure is in principal 1 minus relative firm specific volatility which
then measures the fraction of volatility explained through market and industry stock return variation. Complete comovement, the absence of idiosyncratic volatility, is illustrated
through a COMVc,t measure of 1. The complete absence of comovement is illustrated
through a COMVc,t measure of 0. Since stock return comovement will be the dependent
variable in a subsequent regression model, it could potentially introduce autocorrelation.
To avoid potential autocorrelation in the residuals, a measure derived from COMVc,t is
calculated. Based on each country’s c individual time series of 60 COMVc,t values, the
measure is obtained from the following regression, an AR(1) process, with COMVc,t as
the raw stock return comovement and months denoted by t :
COMVc,t = αc + βc × COMVc,t −1 + εc,t

(4.7)

4.4 Measures

123

The residual εc,t then is the derived stock market comovement measure CMResc,t which
can be used as the dependent variable in the next steps. Durbin-Watson tests are used in
the following regressions to assess whether significant autocorrelation is still existent in the
residuals after transformation.

4.4.2 News Comovement
News comovement measures the comovement of daily news, equivalent to the definition
of stock market comovement. It is comparably based on a decomposition of news volatility into market wide news volatility, industry specific news volatility, and firm specific
news volatility. If higher firm specific volatility is recorded, a higher amount of public
information that is not specific to an industry or entire market should be disseminated. Instead of using excess returns, the decomposition uses the daily news variable as specified in
Section 4.3.2: the daily ‘sentrel’ measure (daily relevance weighted average sentiment of the
news messages of a particular firm) and daily ‘sentiment’ measure. With a high measure of
comovement the public information flow consists of relatively little firm specific information relative to the overall public information flow while low news comovement implies a
high public information flow for specific firms relative to the overall flow of information.
First, daily weighted market news variables and daily weighted industry news variables
have to be computed. Weights wi,c,s and w j ,i,c,s are the same as in the previous section.
Let N j ,i,c,s denote an individual firm’s j daily s news variable (‘sentrel’ and ‘sentiment’) in
country c and industry i then industry news variables are defined as
Ni ,c,s =

X

w j ,i,c,s N j ,i,c,s .

(4.8)

j ∈i

Market news, and as such also country news, variables are defined as
Nc,s =

X

wi,c,s Ni,c,s .

(4.9)

i∈c

Based on the definition of daily market wide, industry specific, and firm specific news variables, three news volatility components are estimated on a monthly basis for each country.
Let Nc,µ be the mean of the weighted market news variable for country c over the entire
sample period, then MNVc,t , the markt wide news volatility of country c in month t , is

124

4 Comovement in International Equity Markets and Public Information

computed as
MNVc,t =

X€

Nc,s − Ncµ

Š2

.

(4.10)

s∈t

Industry specific news volatility INVc,t of country c in month t is defined as
INVc,t =

X

wi,c,s

X€

Ni,c,s − Nc,s

Š2

!
.

(4.11)

s∈t

i ∈c

Firm specific news volatility FNVc,t of country c in month t is defined as

FNVc,t


!
X
X€
X
Š2 
=
w j ,i,c,s
N j ,i ,c,s − Ni,c,s
wi,c,s
.
i ∈c

j ∈i

(4.12)

s∈t

Using above equations, news comovement for a country c in month t is computed as
NCMVc,t = 1 −

FNVc,t
MNVc,t + INVc,t + FNVc,t

.

(4.13)

NCMVc,t is the fraction of news variability that cannot be explained by market or industry wide news. If news exhibits high variability it should also contain some new information, if it exhibits high firm specific variability it should contain some new firm specific
information.

4.5 Results and Interpretation
4.5.1 Descriptive Statistics
For each country, 60 monthly stock market comovement COMV values and 60 news comovement values NCMV are observed in the observation period from 2005 to 2009 which
provides overall 1,380 observation for the entire panel. For CMRes, only 59 observations
per country are available since the measure is constructed with a lagged variable, January
2005 is missing. An overview of all stock return comovement and news comovement measures is presented in Table 4.3. I find, consistent with existing literature (e.g. Morck et al.,
2000; Brockman et al., 2010), the lowest stock return comovement of 0.460 in the United
States followed by the United Kingdom. A comovement of 0.460 implies that 46% of
return fluctuation is common to stocks in the US sample. The United States stock mar-

4.5 Results and Interpretation

125

ket is probably the most developed financial market in the world which should lead to
low comovement. The overall equal weighted mean of stock return comovement is 0.778.
The highest stock market comovement is observed in Denmark, followed by Portugal.
Portugal is also the country with the least number of firms among all countries. Interestingly, all stock return comovements below 0.7 are found in common law countries (see also
Table 4.7). The overall median is quite close to the overall mean with 0.801. By design, the
mean of CMRes should be 0 which is also the case. However, the absolute level of stock
market comovement is not important to the analysis of the time-variation of comovement
in this chapter.
The lowest mean for news comovement can again be found in the United States with
0.572. The highest mean for news comovement is computed from Argentinian public information flow. This can be driven by two slightly distinct factors. First, only a little
amount of public information arrives at the market which also implies that little firm specific public information arrives. Or second, if public information arrives it is not firm specific but industry or market wide information. Still, the absolute level of comovement tells
nothing about how stock market comovement interacts with a variation of news comovement over time. The overall mean of news comovement is, like stock return comovement,
0.778 with a median of 0.800 which is again close to the mean.
Table 4.4 presents regression results of country specific comovement COMV on world
comovement WCMV for each country individually with robust Newey and West (1987)
standard errors. World comovement is calculated in three different ways. First, as the
equal weighted average of country comovements, second, as the trading volume weighted
average of country comovements, and third, as the WFE domestic market capitalization
weighted average of country comovements. For all countries but Portugal11 – whose coefficient is insignificant – a positive, and mostly significant, coefficient for the variation
with world comovement can be observed. Country comovements fluctuate more than
the equal weighted world comovement and mostly more than the volume and market
capitalization weighted world comovement. Consistent with existing literature (Guo and
Savickas, 2008), a common global comovement correlation seems to exist. Time fixed effects in subsequent regressions remove such global trends since the analysis in this chapter
focuses on the time-varying relation of stock return comovement and news comovement.

11

Excluding Portugal from the analyses in this chapter does not change any results.

Overall

Argentina
Australia
Austria
Canada
Denmark
France
Germany
Greece
Hongkong (China)
India
Indonesia
Ireland
Italy
Netherlands
New Zealand
Norway
Portugal
Singapore
Spain
Sweden
Switzerland
United Kingdom
United States

Country

Mean
0.904
0.603
0.874
0.651
0.922
0.749
0.827
0.760
0.776
0.695
0.775
0.847
0.841
0.886
0.776
0.813
0.908
0.723
0.835
0.859
0.818
0.601
0.460
0.778

#Obs
60
60
60
60
60
60
60
60
60
60
60
60
60
60
60
60
60
60
60
60
60
60
60
1,380

0.801

0.908
0.595
0.875
0.645
0.934
0.755
0.833
0.754
0.773
0.705
0.778
0.863
0.860
0.891
0.775
0.844
0.919
0.733
0.838
0.863
0.817
0.615
0.443

Median

COMV

0.129

0.052
0.072
0.044
0.079
0.044
0.066
0.051
0.076
0.058
0.077
0.066
0.068
0.066
0.045
0.065
0.084
0.056
0.067
0.063
0.043
0.062
0.064
0.103

Std.Dev.

0.000

0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000
0.000

Mean

0.006

0.011
-0.002
0.004
-0.004
0.009
0.009
0.006
0.004
-0.007
0.007
0.003
0.002
0.005
0.009
0.011
0.020
0.015
0.011
0.003
0.003
-0.001
-0.001
-0.007

Median

CMRes

0.057

0.046
0.069
0.044
0.068
0.043
0.051
0.048
0.067
0.049
0.075
0.063
0.057
0.052
0.044
0.058
0.078
0.054
0.064
0.054
0.042
0.052
0.056
0.069

Std.Dev.

0.778

0.919
0.686
0.811
0.643
0.902
0.783
0.864
0.732
0.724
0.758
0.608
0.828
0.892
0.879
0.915
0.790
0.913
0.733
0.789
0.747
0.803
0.601
0.572

Mean

0.800

0.943
0.689
0.828
0.647
0.927
0.785
0.869
0.729
0.736
0.767
0.608
0.843
0.900
0.880
0.921
0.786
0.921
0.746
0.796
0.766
0.805
0.604
0.569

Median

0.133

0.077
0.084
0.088
0.121
0.070
0.057
0.041
0.120
0.096
0.077
0.144
0.112
0.040
0.043
0.049
0.065
0.064
0.090
0.063
0.110
0.062
0.066
0.089

Std.Dev.

NCMV (‘sentrel’)

Table 4.3: Descriptive Statistics Monthly Comovement. This table reports descriptive statistics on the monthly stock market comovement
COMVc,t and NCMVc,t for each country in the sample . Stock market comovements (COMV) and news comovements (NCMV) are computed
for each month over the years 2005 to 2009. Additionally, stock market comovement measures which are the residuals from the per country
regression COMVc,t = αc + βC × COMVc,t −1 + εc,t are reported (CMResc,t ). Measures over all countries are equal weighted.

126
4 Comovement in International Equity Markets and Public Information

4.5 Results and Interpretation

127

Table 4.4: Regression of Country Comovement on World Comovement. This table reports
the relation between individual country stock market comovements (COMV) and average world
comovement (WCMV). World comovement is computed in three different ways. First, world
comovement is calculated as the simple mean of all country comovements. Second, it is calculated as the US dollar trading volume weighted average of country comovements and third, it
is calculated as the entire domestic market capitalization weighted average of individual countries. The table reports coefficients and t-statistics from the following regression per country c:
COMVc,t = αc + ηc × WCMVc,t + εc,t . Standard errors are heteroskedasticity and autocorrelation
consistent Newey and West (1987) standard errors and t-statistics are reported in parantheses. Significance at the 1% level is indicated through an ‘a’. 5% and 10% levels are indicated through a ‘b’
and ‘c’.
Country
Argentina
Australia
Austria
Canada
Denmark
France
Germany
Greece
Hongkong (China)
India
Indonesia
Ireland
Italy
Netherlands
New Zealand
Norway
Portugal
Singapore
Spain
Sweden
Switzerland
United Kingdom
United States
Mean
Median

Not Weighted

Volume Weighted

MCap Weighted

Coeff.

t-stat

Coeff.

t-stat

Coeff.

t-stat

2.377a
2.960a
2.845a
2.499a
2.652 b
2.005a
4.376c
3.267a
1.898a
2.713a
3.383a
3.097a
2.302a
2.022a
9.426
3.203a
-97.610
2.386a
2.090a
2.433a
2.147a
1.957a
2.990a

(3.71)
(6.11)
(3.56)
(10.83)
(2.60)
(9.87)
(1.80)
(3.94)
(8.49)
(5.76)
(3.96)
(6.55)
(7.28)
(4.34)
(1.31)
(4.71)
(-0.10)
(6.49)
(7.23)
(4.21)
(6.67)
(9.10)
(13.64)

1.548 b
2.013a
2.133c
1.256a
1.770c
1.093a
2.685
1.844a
1.003a
1.617a
1.701a
1.288a
1.032a
1.174a
6.354
1.472a
-5.654
1.421a
1.088a
1.591 b
1.210a
0.939a
1.360a

(2.24)
(3.20)
(1.98)
(8.00)
(1.78)
(6.20)
(1.28)
(3.23)
(6.01)
(3.78)
(4.12)
(5.65)
(8.08)
(3.17)
(0.92)
(6.06)
(-0.83)
(4.29)
(5.42)
(2.03)
(4.73)
(9.63)
(12.43)

1.528 b
1.778a
1.848a
1.297a
1.618 b
1.100a
2.911
1.951a
1.065a
1.639a
1.910a
1.527a
1.148a
1.373a
11.916
1.612a
-5.587
1.344a
1.144a
1.785 b
1.197a
0.984a
1.396a

(2.53)
(4.15)
(2.75)
(8.38)
(2.22)
(6.69)
(1.35)
(3.46)
(6.64)
(4.17)
(4.02)
(5.74)
(7.26)
(2.91)
(0.61)
(5.31)
(-0.90)
(5.04)
(5.32)
(2.18)
(5.66)
(10.49)
(23.97)

-1.417
2.499

(5.74)
(5.76)

1.389
1.421

(4.50)
(4.12)

1.673
1.527

(5.22)
(4.17)

128

4 Comovement in International Equity Markets and Public Information

4.5.2 Influence of News Comovement on Stock Market Comovement
To generally assess the influence of news comovement on stock market comovement and
to find their time-varying relation (research question 1), a two-way fixed effects model
with month fixed effects and country fixed effects is applied removing the country specific
components and common time trends. Let M c,t be either the raw stock market comovement COMV or CMRes from the first step regression and c, t denotes a country month
combination then the following two-way fixed effects model emerges:
M c,t = αc,t + δ × NCMVc,t + εc,t

(4.14)

The standard errors for the two-way fixed effects model are clustered standard errors (cf.
Petersen, 2009; Thompson, 2011). I also compute the influence of news comovement on
stock market comovement with three news comovement lags, to check whether there are
also lagged dependencies in addition to the contemporaneous relation, using CMRes as the
dependent variable M c,t :
M c,t = αc,t +

3 €
X

Š
δ−k × NCMVc,t −k + εc,t

(4.15)

k=0

Brockman et al. (2010) also present pooled regressions with additional control variables
like the industry Herfindahl index, firm Herfindahl index, and the number of stocks.
However, most of those variables are insignificant in the two-way fixed effects setting and
the number of stocks does not vary in my data set.
The models are estimated using both the ‘sentrel’ and the simple ‘sentiment’ measures
(cf. Section 4.3.2 of this chapter for details on the variable construction). However, I will
focus on the ‘sentrel’ results, ‘sentiment’ results are reported for robustness only and yield
qualitatively the same results. Table 4.5 reports the main results while Table 4.6 reports
only the ‘sentiment’ results; these results are not used in the further discussion.
Table 4.5 reports three models. Model A is based on the normal comovement measure
COMV, Model B uses the derived CMRes measure, and Model C introduces three lags.
The most basic regression, Model A, exhibits a highly significant coefficient of 0.133 with a
t-value of 4.804 which indicates that there is a significant time-varying association between
how firm specific news is dissiminated and how stocks in equity markets comove. The
F-test for fixed effects significantly rejects the null hypothesis of no-fixed effects justifying

4.5 Results and Interpretation

129

the usage of a two-way fixed effects model. The Hausman test as well as the BreuschPagan Lagrange Multiplier test for random effects significantly reject the null hypothesis of
random effects which in combination with the F-test again justifies the fixed effects model.
R2 is very high at 84.71%, probably strongly driven through the fixed effects. DurbinWatson statistics indicate strong positive autocorrelation in the residuals which is why I
resort to the CMRes measure that has the first order autocorrelation of COMV removed.
Durbin-Watson statistics are calculated for each country over the time-series and mean as
well as median values are presented in Table 4.5.
Model B presents the contemporaneous association of news comovement with CMRes.
However, I lose all January 2005 observations for the panel since the derivation of CMRes
requires a lagged variable. The coefficient for NCMV is still highly significant with a value
of 0.094 for the estimate. Tests for no-fixed effects and random effects still significantly
reject the null hypotheses. The two-way fixed effects model is also suitable for the model
based on CMRes. Adjusted R2 is significantly reduced in comparison to using COMV as
the dependent variable. However, one must not be tempted to compare the R2 measures.
Deriving CMRes from COMV greatly reduces fixed effects which drives a reduction in
adjusted R2 and it is by far no sign of Model B being worse than Model A. Durbin-Watson
statistics for Model B show that there is no significant autocorrelation left in the residuals.
In addition to the purly contemporaneous regressions, I also introduce Model C with
three lags. By design, this again reduces the number of available months per country.
The contemporaneous coefficient is still significant, positive, and almost the same as for
Model B. Lags one through three are all not significant and the effect of the contemporaneous term is not mitigated. Since lags do not add explanatory value to the model, further
regressions focus on the contemporaneous association between stock return comovement
and news comovement.
Economically, a significant positive coefficient confirms my hypothesis that a part
of stock return comovement is driven by time-varying information production proxied
through RNSE firm specific newswire messages. These results confirm Brockman et al.
(2010) who also find that information production has a significant impact on stock return
comovement. Their analysis is based on a lower frequency than mine which suggests that
there are influences of information production on stock return comovement on different
frequencies. Brockman et al. (2010) ultimately relate their stock return comovement measures to business cycles which are measured on a three months frequency. The novelty
of my results, in contrast to existing literature, is that I am able to directly relate a major

130

4 Comovement in International Equity Markets and Public Information

source of information production with stock return comovement. The results in Table 4.5
show that a relatively high flow in public firm specific information reduces comovement.
The coefficient is positive in the regression since high news comovement implies relatively
little firm specific information. Two cases for high news comovement can generally be
distinguished, it might be the case that no public information arrives or that only market
and industry wide information is disseminated.
I show in Chapter 2 that market participants obtain private information from public
information sources, potentially through analysis and interpretation. Market participants
have limited cognition and limited research resources, a fact that can drive informed trading
even if information is public. Some market participants are better than others in processing firm specific public information while others again are just slow to observe that firm
specific information has even arrived. The reduction in comovement due to a higher relative amount of firm specific news is then potentially a result of derived private information
that is capitalized into stock returns. The capitalization of private information increases
idiosyncratic variability of stock prices which in turn reduces stock return comovement
(Durnev et al., 2003).
In addition, a relatively high firm specific public information flow might incentivize
market participants to obtain additional private information. This, in turn, enhances the
comovement reducing effect of firm specific public information and increases idiosyncratic
stock volatility. A higher firm specific information flow enhances the firm specific information environment making a firm more transparent to the market. A trader has limited
time and intellectual capacity to observe stocks and profitable trading opportunities. Additional firm specific public information might lead a certain fraction of traders to the
conclusion that it could be interesting to trade in a certain stock. If they generate additional private information through research, analysis, and purchasing information this
private information is eventually priced into the stock price once they trade on their information. This trading then increases a firm’s idiosyncratic volatility and reduces stock
market comovement.
My results are also consistent with the theoretical model of Veldkamp (2006) which predicts that a relatively little flow of firm specific information increases comovement and
higher information production decreases stock return comovement. In general, prices
should be more efficient when more information is capitalized into individual stocks
(Durnev et al., 2004). Results from Table 4.5 provide evidence for the link between a time
variation of stock market comovement and the flow of firm specific information.

4.5 Results and Interpretation

131

However, this analysis focuses on the overall panel of countries and not subsamples
based on country characteristics. The next section investigates whether there are crosssectional characteristics that influence the association between news comovement and
stock return comovement (research question 2).

4.5.3 Cross-Country Analysis
Existing research shows that stock return comovement varies substantially between countries (cf. Jin and Myers, 2006; Morck et al., 2000; Karolyi et al., 2009). Since country
characteristics influence stock return comovement, it is reasonable to assume that the association of news comovement and stock returns might also be influenced by country
characteristics, e.g. transparency or corruption.
Table 4.7 provides a descriptive overview on additional country specific information
such as information about a country’s legal system, corruption, ICT development, per
capita GDP, antidirector rights, accounting quality, and whether a country is a developing
country or not. The detailed description of those variables is available in Section 4.3.3 of
this chapter. The sample of 23 countries comprises of 9 common law countries and 14
civil law countries which describes countries with either a French, German, or Scandinavian legal tradition. The Corruption Perceptions Index (CPI) is by far the lowest for the
three developing countries in my sample, Argentina, Indonesia, and India. The countries
with the lowest perceived corruption (highest index values) are Denmark, New Zealand,
Singapore, and Sweden. On average, the index is on the level of the United States or Germany which shows that considering the cross-section of countries that exist in the world,
the sample comprises relatively few overly corrupt countries. The ICT development index
shows two clear outliers with Indonesia and India both having on average underdeveloped
information and communication systems. This might seem strange for India since it is one
of the major countries to which the US outsources call centers, software development, or
even administrative medical work. However, one must keep in mind that India still has a
huge rural population without access to communication or information systems. Again,
per capita GDP is by far the lowest for the three developing countries in the sample. Norway has the highest per capita GDP with an average of 78,705 US dollars over the years
2005 to 2009, largely driven by their immense oil and gas resources.

4 Comovement in International Equity Markets and Public Information

132

Table 4.5: Influence of News Comovement on Stock Market Comovement. This table presents
regression results for the influence of news comovement on stock market comovement. Three
regression models are provided. Model A is the naive approach, regressing the raw stock market comovement COMV on news comovement NCMV. In model B the residuals of the country specific
regressions COMVc,t = αc + βc × COMVc,t −1 + εc,t (CMRes) are regressed on contemporaneous news comovement. Model C adds three lags to model B. Regressions are two-way fixed effects
models over all countries and all month; models with lags naturally lose observations. Adjusted R2
and additional statistics to assess the two-way fixed effects model are provided. The F-test tests for
no-fixed effects while the Hausman and Breusch-Pagan Lagrange Multiplier tests test for random
effects. Mean and median values of the Durbin-Watson statistics for per country regressions without fixed effects are also provided. Robust t-statistics are reported in parantheses. Significance at
the 1% level ist denoted by an ‘a’.
NCMV
Coeff.
t-stat

Model A (COMV)

Model B (CMRes)

Model C (CMRes)

0.133a
(4.804)

0.094a
(3.804)

0.091a
(3.357)

NCMV l a g 1
Coeff.
t-stat

-0.010
(-0.333)

NCMV l a g 2
Coeff.
t-stat

0.023
(1.446)

NCMV l a g 3
Coeff.
t-stat

0.005
(0.210)

Number of Observations

1,380

1,357

1,311

84.71%

29.64%

30.25%

F-Test (No FE)
F-stat
p-value

49.60
< 0.0001

6.52
< 0.0001

6.58
< 0.0001

Hausman Test (RE)
m-stat
p-value

27.86
< 0.0001

18.38
< 0.0001

12.44
0.0060

Breusch-Pagan LM Test (RE)
m-stat
p-value

9,188
< 0.0001

915
< 0.0001

938
< 0.0001

1.310
1.297

2.111
2.101

2.118
2.099

0.058
0.002

0.636
0.649

0.632
0.636

0.942
0.998

0.365
0.351

0.368
0.364

Adj. R2

DW Statistics
Mean
Median
DW p-value (Pr < DW)
Mean
Median
DW p-value (Pr > DW)
Mean
Median

4.5 Results and Interpretation

133

Table 4.6: Influence of News Comovement on Stock Market Comovement – ‘Sentiment’ Only.
This table presents regression results for the influence of news comovement on stock market comovement. To check robustness, news comovement in this table is based on the sentiment measure
only. Three regression models are provided. Model A is the naive approach, regressing the raw
stock market comovement COMV on news comovement NCMV. In model B the residuals of regressions COMVc,t = αc + βc × COMVc,t −1 + εc,t (CMRes) are regressed on contemporaneous
news comovement. Model C adds three lags to model B. Regressions are two-way fixed effects
models over all countries and all month; models with lags naturally lose observations. Adjusted R2
and additional statistics to assess the two-way fixed effects model are provided. The F-test tests for
no-fixed effects while the Hausman and Breusch-Pagan Lagrange Multiplier tests test for random
effects. Mean and median values of the Durbin-Watson statistics for per country regressions without fixed effects are also provided. Robust t-statistics are reported in parantheses. Significance at
the 1% level ist denoted by an ‘a’.
NCMV
Coeff.
t-stat

Model A (COMV)

Model B (CMRes)

Model C (CMRes)

0.159a
(3.829)

0.101a
(3.001)

0.093a
(2.746)

NCMV l a g 1
Coeff.
t-stat

-0.027
(-1.443)

NCMV l a g 2
Coeff.
t-stat

0.042
(1.426)

NCMV l a g 3
Coeff.
t-stat

0.025
(0.780)

Number of Observations

1,380

1,357

1,311

84.60%

29.11%

29.84%

F-Test (No FE)
F-stat
p-value

44.21
< 0.0001

6.42
< 0.0001

6.54
< 0.0001

Hausman Test (RE)
m-stat
p-value

35.58
< 0.0001

13.65
0.0002

10.58
0.0142

Breusch-Pagan LM Test (RE)
m-stat
p-value

8,091
< 0.0001

911
< 0.0001

935
< 0.0001

1.334
1.312

2.123
2.114

2.112
2.097

0.050
0.003

0.635
0.655

0.618
0.629

0.950
0.997

0.365
0.345

0.382
0.371

Adj. R2

DW Statistics
Mean
Median
DW p-value (Pr < DW)
Mean
Median
DW p-value (Pr > DW)
Mean
Median

Mean
Median

Argentina
Australia
Austria
Canada
Denmark
France
Germany
Greece
Hongkong (China)
India
Indonesia
Ireland
Italy
Netherlands
New Zealand
Norway
Portugal
Singapore
Spain
Sweden
Switzerland
United Kingdom
United States

Country
4.12
6.58
6.32
6.34
7.22
6.16
6.61
5.25
6.70
1.59
2.13
6.37
6.18
7.14
6.44
7.09
5.47
6.57
5.91
7.50
6.94
6.78
6.44
5.99
6.44

2.9
8.6
8.1
8.7
9.4
7.3
7.8
4.6
8.3
3.5
2.3
7.5
5.2
9.0
9.4
8.7
6.5
9.3
6.7
9.3
9.0
8.4
7.2
7.29
8.10

Civil Law
Common Law
Civil Law
Common Law
Civil Law
Civil Law
Civil Law
Civil Law
Common Law
Common Law
Civil Law
Common Law
Civil Law
Civil Law
Common Law
Civil Law
Civil Law
Common Law
Civil Law
Civil Law
Civil Law
Common Law
Common Law

ICT Development
Index

Corruption
Perceptions Index

Legal
System

36,420
39,248

6,550
40,354
43,157
40,407
54,592
39,248
39,003
26,733
28,638
983
1,893
54,269
34,211
45,817
28,056
78,705
20,282
35,474
30,582
46,422
56,315
40,507
45,460

Per Capita GDP
(US Dollars)

3.30
4.00

4
4
2
5
2
1
2
4
3
5
2
5
2
5
4
1
2
4
4
3
4
3
5

Antidirector
Rights

65.24
68.00

45
75
54
74
62
69
62
55
69
57
n/a
n/a
62
64
70
74
36
78
64
83
68
78
71

Accounting
Quality

Yes
Yes

Yes

Developing
Country

Table 4.7: Descriptive Statistics Countries. This table presents descriptive statistics for each country. Legal System indicates whether a
country is a common law or civil law country. The Corruption Perceptions Index is based on the year 2007; a higher number indicates
lower perceived corruption in a country. The United Nations ICT Development Index represents a higher standard in information and
communication technology the higher the number. Per capita GDP is based on the average over the years 2005 to 2009 and reported in US
dollars. Antidirector rights and accounting quality are derived from Andrei Shleifer’s data set. The potential range for antidirector rights is 1
to 6 while a higher number indicates stronger antidirector rights. The accounting quality index represents a higher accounting quality with
higher index numbers. The classification into a developing country is taken from the Worldbank classification.

134
4 Comovement in International Equity Markets and Public Information

4.5 Results and Interpretation

135

Both variables for antidirector rights and accounting quality are based on Andrei
Shleifer’s data sets. The least antidirector rights are granted to investors in France and
Norway, both civil law countries. Most rights are given to investors in the United States,
Canada, India, Ireland, and the Netherlands, all but the Netherlands common law countries. The theoretical highest value of 6 is reached by non of the countries in my sample.
The highest accounting quality can be found in Sweden and the lowest in Portugal. Accounting quality data for Ireland and Indonesia are missing.
To assess whether country characteristics influence the association between stock return
comovement and news comovement, different subsamples based on country and market
(stock market price and volume data based) specific criteria are compiled. Regression models are estimated for a subset of data exactly comparable to Equation 4.14. Only Model B
is estimated for country subsamples which implies that the dependent variable is always
CMRes. Depending on country characteristics, the flow of firm specific information might
have different magnitudes of influence on stock return comovement and thus also on the
idiosyncratic variability of stock prices. Tables 4.8 and 4.9 present the results for different subsamples. Table 4.8 presents subsample estimations based on the entire domestic
market capitalization, the per firm trading volume in US dollars in the sample, whether
a country is a civil law country or not, and perceived corruption in a country. Market
capitalization and per firm value are both based on market criteria while civil law country
and corruption concern the entire institutional setting of one country. Table 4.9 presents
results for measures that focus more on economic indicators and institutional settings: the
per capita GDP, the strength of antidirector rights, and accounting quality. Specifically, the
legal tradition and corruption as well as investors’ rights against management and a firm’s
transparency measured through its accounting quality potentially have a significant direct
impact on the overall information environment of firms. Tables 4.8 and 4.9 additionally
provide information on the number of observations, adjusted R2 , and tests for no-fixed
effects and random effects.
The division into subsamples by domestic market capitalization and per firm trading
volume yields very similar results. Those countries with smaller domestic market capitalization and a lower per firm trading volume have a highly significant association of contemporaneous news comovement NCMV with stock return comovement having coefficients
of 0.143 and 0.136 respectively. The t-values of both regressions are highly significant at
4.706 and 4.616. Coefficients are not statistically significant for both market capitalization and per firm trading volume in countries with high domestic market capitalization

136

4 Comovement in International Equity Markets and Public Information

and high per firm trading volume. Although not significant, both coefficients still have
the expected sign. This is consistent with the fact that larger firms are more transparent
to outside investors (Bushman et al., 2004). If large firms are more transparent it might
be that additional firm specific information does not add much to existing firm specific
information which is already at a high level. Also, such information is more likely to be
already capitalized into individual stock prices. But if additional firm specific information
does not add much to existing firm specific information, the relation between the flow of
public firm specific information and stock return comovement or idiosyncratic variability
of stock prices should be low. This is exactly what I find in the subsamples with high domestic market capitalization and high per firm trading volume. If transparency is lower
on the other hand, firm specific public information might add much more to the firm
specific information set and also incentivizes other traders to obtain private information.
Such behavior and information characteristics can then lead to a stronger association of
news comovement with stock return comovement as observed for countries with small
domestic market capitalization and small per firm trading volume.
The legal tradition of a country, common law or civil law, is an important characteristic
for the economic environment of both firms and external investors. Dividing the sample
of 23 countries into civil law countries and common law countries, unfortunately results
in two samples which are of different size. The subsamples contain 826 observations for
civil law countries and only 531 for common law countries. Civil law countries have a
positive and significant coefficient of 0.128 for the relation of stock return comovement
with news comovement. The coefficient for common law countries is only 0.048 and not
statistically significant. However, I cannot asses whether the statistical insignificance for
common law countries is not partially driven by the smaller sample size. Bushman et al.
(2004) find higher corporate governance transparency in common law countries in comparison to civil law countries. In addition, previous research finds on average lower stock
return comovement in common law countries (Khandaker and Heaney, 2009) which the
comovement descriptive statistics in this chapter confirm (cf. Table 4.3). Thus, the difference in the association of stock return comovement and the comovement of firm specific
news might be driven by two factors in the civil law and common law subsamples, transparency and the prevailing average level of stock return comovement. As in the previous
paragraph, higher transparency might have a decreasing effect on the association of firm
specific information flow with stock return comovement. If the prevailing level of idiosyncratic volatility is already high, as in common law countries, prices might already comprise

4.5 Results and Interpretation

137

more firm specific information which gives less leeway for the idiosyncratic variability of
stock prices to increase. This results in a lower average association of news comovement
with stock market comovement in common law countries.
One measure that is related to transparency and also defines a firm’s general information
environment is corruption, an important institutional feature of a country. I find that in
countries with more corruption the association between stock return comovement and
news comovement is highly significant with a coefficient of 0.130 and a t-value of 5.610.
It is not significant for the less corrupt half of countries in my sample. However, in comparison to the differentiation by market capitalization and per firm trading volume the
coefficient is very close to being significant. Karolyi et al. (2009) and Li et al. (2004) include corruption in their good government measures. In combination, the rule of law and
freedom from corruption have a decreasing effect on stock return comovement in their
studies. Again, the same explanation as for the paragraphs above applies. Lower corruption potentially increases transparency to investors and enhances the general information
environment in a country. Lower transparency increases the effect that additional firm
specific information has on the association between stock return comovement and news
comovement. The same effect is found for subsamples constructed on the per capita GDP
in US dollars (cf. Table 4.9) consistent with existing literature which also adresses transparency as a determinant of stock return comovement in less developed countries (Karolyi
et al., 2009). The coefficient of the half of the sample with lower per capita GDP, which
still are mostly developed countries, is 0.139 and highly statistically significant. Although
positive, the coefficient is only 0.056 and not significant for the countries in the sample
with higher per capita GDP.
The separation of subsamples in Table 4.9, by antidirector rights and accounting quality, focuses on certain specific aspects that influence a firm’s information environment and
the disposition of investors to acquire firm specific information. In contrast to all previous subsample pairs, both, the subsamples for antidirector rights as well as for accounting
quality, do not exhibit such clear cut differences. The coefficient for countries with less
antidirector rights is 0.103 and highly significant at the 1% level while the coefficient for
countries with more antidirector rights is only 0.074 while being still significant at the
10% level. Existing literature confirms that lower porperty rights disincentivize investors
to obtain private firm specific information (Morck et al., 2000). In turn, this might lead to
a higher impact of firm specific news on idiosyncratic stock price variability. Once firm
specific information arrives, it still needs to be capitalized into stock prices since the in-

138

4 Comovement in International Equity Markets and Public Information

formation might not have been obtained as firm specific private information by investors
before. The specific characteristic of accounting quality as a country specific institutional
variable directly influences a firm’s transparency and is linked to different legal systems.
For the half of countries with the lower accounting quality, the coefficient for the association of news comovement with stock return comovement is 0.144 and highly significant.
The coefficient for countries with higher accounting quality is much lower at 0.069 but
still significant at the 10% level. Again, an explanation for this behavior is the influence of
transparency.
Summarizing the subsample results, it becomes clear that different institutional settings,
characterized through a variety of country parameters, considerably influence the association between stock return comovement and news comovement. The results are driven
by two comprehensive factors: a firm’s information environment and the legal protection
of investors which indirectly also influences firm specific information in stock prices. In
more opaque stock markets, stock markets where firms are less transparent to outside investors, additional firm specific information can have a stronger influence on firm specific
stock price volatility. Since less information is capitalized into stock prices on a base level,
new firm specific information potentially has a stronger effect on idiosyncratic volatility.
Also, in opaque markets it is more likely that firm specific news still contains information
that has not yet been found by outside investors. Direct effects of a disadvantageous information environment are potentially amplified through lower investor protection that
disincentivizes private information gathering. News specifically seems to enhance the efficieny of stock prices in an environment where it is legally as well as economically more
difficult for outside investors to obtain firm specific information.

Breusch-Pagan LM Test (RE)
m-stat
p-value

Hausman Test (RE)
m-stat
p-value

F-Test (No FE)
F-stat
p-value

Adj. R2

Number of Observations

NCMV
Coeff.
t-stat

72.84
< 0.0001

12.40
0.0004

2.84
< 0.0001

25.75%

708

0.143a
(4.706)

Lower Half

467.46
< 0.0001

0.52
0.4730

6.54
< 0.0001

43.46%

649

0.024
(0.711)

Upper Half

Domestic MCap

122.08
< 0.0001

14.34
0.0002

3.41
< 0.0001

28.29%

708

0.136a
(4.616)

Lower Half

381.44
< 0.0001

0.53
0.4686

5.71
< 0.0001

40.18%

649

0.025
(0.768)

Upper Half

Per Firm Volume

223.63
< 0.0001

13.87
0.0002

4.11
< 0.0001

29.51%

826

0.128a
(4.910)

Yes

257.54
< 0.0001

1.79
0.1814

5.09
< 0.0001

42.18%

531

0.048
(1.209)

No

Civil Law Country

136.90
< 0.0001

12.73
0.0004

3.52
< 0.0001

28.98%

708

0.130a
(5.610)

More Corrupt

308.04
< 0.0001

3.59
0.0581

5.06
< 0.0001

37.39%

649

0.059
(1.537)

Less Corrupt

Corruption (CPI)

Table 4.8: Influence of News Comovement on Stock Market Comovement – Subsamples 1. This table presents regression results for the
influence of news comovement on stock market comovement for subsamples of the 23 countries. Subsamples are constructed based on the
fact whether countries have a civil law legal system or not, whether countries belong to the lower half or upper half of the entire domestic
market capitalization, based on the average per firm trading volume, and on whether countries belong to the more or less corrupt half in the
sample. Stock market size is the entire domestic market capitalization averaged over the years 2005 to 2009 in US dollars based on data from
the World Federation of Exchanges. Average per firm trading volume is average yearly per firm trading volume in US dollars over the years
2005 to 2009. Corruption is based on the the Corruptions Perceptions Index from the year 2007. In all subsample regressions the residuals of
the country specific regressions COMVc,t = αc + βc × COMVc,t −1 + εc,t (CMRes) are regressed on contemporaneous news comovement
(NCMV). Regressions are two-way fixed effects models over all countries and all months; the first observation of each country’s time series is
lost as a result of the specification of the dependent variable. Adjusted R2 and additional statistics to assess the two-way fixed effects model are
provided. The F-test tests for no-fixed effects while the Hausman and Breusch-Pagan Lagrange Multiplier tests test for random effects. Robust
t-statistics are reported in parantheses. Significance at the 1% level ist denoted by an ‘a’.

4.5 Results and Interpretation
139

Breusch-Pagan LM Test (RE)
m-stat
p-value

Hausman Test (RE)
m-stat
p-value

F-Test (No FE)
F-stat
p-value

Adj. R2

Number of Observations

NCMV
Coeff.
t-stat

145.63
< 0.0001

13.21
0.0003

3.62
< 0.0001

29.56%

708

0.139a
(4.481)

Lower Half

293.88
< 0.0001

3.26
0.0708

4.92
< 0.0001

36.78%

649

0.056
(1.590)

Upper Half

Per Capita GDP

129.26
< 0.0001

10.85
0.0010

3.38
< 0.0001

28.87%

649

0.103a
(3.244)

Lower Half

286.42
< 0.0001

4.52
0.0335

4.72
< 0.0001

34.55%

708

0.074c
(1.836)

Upper Half

Antidirector Rights

120.80
< 0.0001

8.24
0.0041

3.36
< 0.0001

30.38%

649

0.144a
(4.430)

Lower Half

342.37
< 0.0001

4.30
0.0381

5.77
< 0.0001

42.69%

590

0.069c
(1.762)

Upper Half

Accounting Quality

Table 4.9: Influence of News Comovement on Stock Market Comovement – Subsamples 2. This table presents regression results for the
influence of news comovement on stock market comovement for subsamples of the 23 countries. Subsamples are constructed on the per capita
GDP in US dollars, the antidirector rights index, and the accounting quality index. Lower half in this tables indicates, the half of the countries
with the lower per capita GDP, less antidirector rights, and less transparent accounting standards. No accounting quality measures exist for
Indonesia and Ireland such that for the accounting quality regression only 21 countries are taken into account. In all subsample regressions
the residuals of the country specific regressions COMVc,t = αc + βc × COMVc,t −1 + εc,t (CMRes) are regressed on contemporaneous news
comovement (NCMV). Regressions are two-way fixed effects models over all countries and all months; the first observation of each country’s
time series is lost as a result of the specification of the dependent variable. Adjusted R2 and additional statistics to assess the two-way fixed
effects model are provided. The F-test tests for no-fixed effects while the Hausman and Breusch-Pagan Lagrange Multiplier tests test for random
effects. Robust t-statistics are reported in parantheses. Significance at the 1% level ist denoted by an ‘a’, ‘b’ at the 5% level, and ‘c’ at the 10%
level.

140
4 Comovement in International Equity Markets and Public Information

4.5 Results and Interpretation

141

In addition to constructing subsamples, I provide correlations of news comovement,
the association of news comovement with CMRes, and institutional variables for individual countries. One has to keep in mind that the sample for the correlation coefficient is
quite small with only 23 observations and only 21 for accounting quality. Correlation
coefficients are Spearman’s rank correlation coefficients (cf. Brockman et al., 2010). To
calculate country specific associations, I compute for each country c and month t
CMResc,t = αc + γc × NCMVc,t + εc,t

(4.16)

resulting in 23 individual γc s. However, one caveat in comparison to the panel regressions
remains, I cannot incorporate monthly time dummies. Results of the correlation analysis are presented in Table 4.10. Variables that correlate significantly with the average per
country news comovement are the civil law variable, domestic stock market size, and accounting quality. A civil law country is more likely to have high news comovement while a
large stock market and higher accounting quality relate to lower news comovement. Those
observations are consistent with the descriptive statistics presented earlier in this chapter.
The country specific association of news comovement with stock market comovement is
significantly negatively related to stock market size and accounting quality, consistent with
subsample results. ICT development is highly negatively correlated with corruption and
has a high positive correlation with per capita GDP. A negative and significant correlation
coefficient shows that civil law countries have on average less antidirector rights than common law countries. Interestingly, accounting quality is significantly correlated with all
variables but antidirector rights. All correlation coefficients for accounting quality exhibit
the expected direction. Accounting quality is higher in more developed countries with
large stock markets, it is higher in countries with a common law legal tradition, with high
ICT developement, and low corruption. Countries with high accounting quality have a
significantly lower association of news comovement with stock return comovement. Interestingly, per capita GDP has a correlation with a country’s legal tradition that is close
to zero.
In general, the significant correlation results are consistent with the subsample analyses.
The insignificant correlation of accounting quality and antidirector rights shows that not
all characteristics which are used to build subsamples are necessarily highly correlated but
nonetheless have explanatory power. Different characteristics of a country’s institutional
setting can influence the association of news comovement and stock return comovement.

-0.202
(0.356)

Association of NCMV with CMRes
(CMResc,t = αc + γc × NCMVc,t + εc,t )

Antidirector Rights
(1-6, 6 = strongest rights)

Per Capita GDP
(US Dollars)

Stock Market Size
(Domestic MCap)

Civil Law Country
(1 = yes, 0 = no)

ICT Development Index
(higher = more developed)

Corruption Perceptions Index
(0 - 10, 10 = no corruption)

0.080
(0.716)

Corruption
Perceptions Index

News Comovement
(Country Mean)

N = 23
(Accounting, N = 21)

-0.040
(0.855)

-0.195
(0.373)

0.843a
(< 0.001)

-0.336
(0.117)

0.176
(0.422)

-0.012
(0.957)

0.130
(0.553)

-0.029
0.897

0.187
(0.392)

0.126
(0.588)

0.475 b
0.030

0.370c
(0.098)

-0.520 b
(0.016)

-0.567a
(0.005)
0.067
(0.761)

0.578a
0.006
-0.084
0.703

0.783a
(< 0.001)

0.617a
(0.003)

-0.476 b
(0.030)

- 0.617 b
(0.003)

Accounting
Quality

0.037
(0.868)

-0.060
(0.787)

-0.193
(0.378)

Antidirector
Rights

0.642a
(0.001)

-0.194
(0.376)

-0.522 b
(0.011)

0.269
(0.215)

-0.018
(0.936)

-0.612a
(0.002)

0.457 b
(0.029)

Per Capita GDP
(US Dollars)

Stock Market Size
(Domestic MCap)

Civil Law
Country

-0.279
(0.198)

-0.006
(0.977)

ICT Development
Index

Table 4.10: Cross-Sectional Correlations. This table reports correlations between cross-sectional characteristics of countries. News Comovement is the average news comovement in a country. The association of news comovement with CMRes is the coefficient of the country c
specific regressions CMResc,t = αc + γc × NCMVc,t + εc,t . The Corruptions Perceptions Index is based on the year 2007; a higher number
indicates lower perceived corruption in this country. The United Nations ICT Development Index represents a higher standard in information and communication technology the higher the number. ‘Civil Law Country’ indicates whether a country is a civil law country or not.
‘Stock Market Size’ is the entire domestic market capitalization averaged over the years 2005 to 2009 in US dollars based on data from the
World Federation of Exchanges. Per capita GDP is based on the average over the years 2005 to 2009 and reported in US dollars. Antidirector
rights and accounting quality are derived from Andrei Shleifer’s data set. The potential range for anti-director rights is 1 to 6 while a higher
number indicates stronger antidirector rights. The accounting quality index represents a higher accounting quality with higher index numbers.
Correlation coefficients are Spearman’s rank correlation coefficients and p-values are reported in parantheses. Significance at the 1% level ist
indicated through an ‘a’. 5% and 10% levels are indicated through ‘b’ and ‘c’.

142
4 Comovement in International Equity Markets and Public Information

Breusch-Pagan LM Test (RE)
m-stat
p-value

Hausman Test (RE)
m-stat
p-value

F-Test (No FE)
F-stat
p-value

Adj. R2

Number of Observations

NCMV
Coeff.
t-stat

Number of Stocks

572.38
< 0.0001

0.03
0.8605

5.12
< 0.0001

26.69%

1,180

0.013
(0.938)

3134

Overall

309.80
< 0.0001

0.19
0.6601

13.04
< 0.0001

77.85%

295

0.022 b
(2.265)

782

Quantiles 1-5

186.73
< 0.0001

0.00
0.9638

6.83
< 0.0001

64.78%

295

0.009
(0.647)

785

Quantiles 6-10

60.38
< 0.0001

0.30
0.5824

3.07
< 0.0001

45.65%

295

0.017
(1.295)

785

Quantiles 11-15

8.31
0.0039

2.15
0.1427

1.47
0.0231

28.61%

295

0.002
(0.044)

782

Quantiles 16-20

Table 4.11: Influence of News Comovement on Stock Market Comovement – USA Only. This table presents regression results for influence
of news comovement on stock market comovement for the United States only. The complete US sample of 3134 stocks is splitted into 20
quantiles by average per firm trading volume. Comovement and news comovement are calculated for each quantile separately. Subsamples
for quantiles 1-5, 5-10, 11-15, and 16-20 are also reported. In the overall sample as well as subsamples the residuals of the country specific
regressions COMVc,t = αc + βc × COMVc,t −1 + εc,t (CMRes) are regressed on contemporaneous news comovement (NCMV). Regressions
are two-way fixed effects models. The first observation of each quantiles’s time series is lost as a result of the specification of the dependent
variable. Adjusted R2 and additional statistics to assess the two-way fixed effects model are provided. The F-test tests for no-fixed effects while
the Hausman and Breusch-Pagan Lagrange Multiplier tests test for random effects. Robust t-statistics are reported in parantheses. Significance
at the 1% level ist denoted by an ‘a’.

4.5 Results and Interpretation
143

144

4 Comovement in International Equity Markets and Public Information

4.5.4 US Specific Analysis
The analysis for only the United States is based on the same methodology as the international analysis in Section 4.5.2 of this chapter. In contrast to the international subsample
analysis, all stocks are traded within the same institutional setting in the United States such
that all country specific variables used in the previous section do not vary. Consequently,
firm specific and not external characteristics drive potential differences. Considering all
countries in the main sample, the United States have a highly developed stock market with
transparent firms and strong investor protection. All 3134 stocks in the US sample are
separated into 20 quantiles based on the average per firm trading volume which enables the
usage of the panel regression methodology. Quantile 1 includes the firms with the largest
trading volume and quantile 20 those with the smallest trading volume. In addition, panel
regressions are also computed on subsamples: quantiles 1-5, quantiles 6-10, quantiles 11-15,
and quantiles 16-20. Table 4.11 presents results for the overall and subsample regressions
including the F-test, Hausman test, and Breusch-Pagan Lagrange Multiplier test statistics.
The coefficient of the overall regression is slightly positive but insignificant. The only coefficient that is found to be significant is for quantiles one to five which includes the 782
stocks with the largest average per firm trading volume. The significant coefficient of quantiles one to five is also very small in comparison to the international analysis. Although
not significant, all coefficients are positive.
Within the US (same institutional setting), Irvine and Pontiff (2009) show that smaller
firms are riskier and thus have a higher idiosyncratic volatility. Larger US firms also enjoy
higher analyst coverage which can increase stock return comovement (Piotroski and Roulstone, 2004). If idiosyncratic volatility is already high it might be the case that additional
firm specific information does not significantly further increase firm specific volatility.
Such characteristics could explain why I only find a significant association of news comovement and stock return comovement for the 782 largest stocks in the US sample. At
first the results may seem a bit contradictory to results from previous sections. However,
it is important to keep in mind that in this analysis, institutional settings along the cross
section do not vary, only firm characteristics vary. In addition, multiple factors influence
the association of stock market comovement and news comovement. Such a design implies
that institutional settings cannot drive differences among quantiles. The important finding is that also for the US stock market alone, all regression coefficients show at least the
expected sign, consistent with results from previous sections.

4.6 Conclusion

145

4.6 Conclusion
In this chapter, I study the effect of the flow of firm specific public information on stock return comovement, thus also firm specific stock price variability, over 23 countries. In contrast to existing literature, a direct measure of firm specific information based on Thomson Reuters newswire messages is applied. The modelling of stock return comovement
is based on Campbell et al. (2001) and Brockman et al. (2010) while I construct a news
comovement measure similar to stock return comovement. Specifically, I am interested
in the time-varying influence of firm specific information on stock return comovement,
overall and in addition separated by country characteristics. The regression framework
uses monthly measures from 2005 to 2009.
Results provide evidence that stock return comovement is linked to the relative amount
of firm specific public information that arrives at a stock market. The relative amount is
defined through the construction of the news comovement measure. If news comovement
decreases, more firm specific information is disseminated relative to market or industry
information. More relative firm specific information reduces stock return comovement
consistent with existing empirical literature that uses proxies for information production
(Brockman et al., 2010) and consistent with theoretical models (Veldkamp, 2006). The firm
specific flow of public information seems to contain information that is not yet capitalized
into stock returns which in turn increases idiosyncratic volatility of stock prices. In addition, a relative increase of firm specific information might incentivize investors to obtain
further private firm specific information amplifying the stock return comovement reducing effect. Country specific institutional characteristics significantly affect the strenght of
the association of news comovement with stock return comovement. The information
environment of firms and investor protection are the major drivers of differences between
countries. More developed financial markets with transparent firms and strong outside
investor protection generally show a lower magnitude of association between the firm specific flow of public information and stock return comovement.
The main contribution of this chapter is that I show that information production, in
contrast to existing literature directly measured through firm specific public information,
significantly influences stock return comovement and thus the efficiency of financial markets. In addition, I find that despite global integrated financial markets, strong differences
in the information processing capabilities of international stock markets remain, also as a
result of external characteristics.

Chapter 5
Conclusion
5.1 Summary
The central message of this thesis is that the firm specific flow of public information has
a significant impact on financial markets. In addition, I find that today’s equity markets
show expeditious reactions to news and as a result speedy information processing. As an
overall research matter, this thesis investigates the effect of firm specific news on equity
markets from three different perspectives. Chapter 2 analyzes high-frequency intraday
market dynamics around the arrival of firm specific news, Chapter 3 focuses on the impact
that firm specific news has on trading in fragmented market and on fragmentation characteristics, while Chapter 4 takes a broader perspective and investigates the association
of firm specific public information and stock return comovement in international equity
markets. The overarching question that motivates this thesis is how information influences
financial markets and how it is incorporated into prices. Understanding those mechanisms
is central to our comprehension of modern financial markets.
Chapter 2 introduces two specific research questions. How do firm specific news messages, separated by their tone, influence intraday market dynamcis, i.e. price discovery,
liquidity, and trading intensity? In addition, one central question is how those market
measures interact around firm specific newswire messages. In contrast to existing literature, I am able to differentiate by the tone of a newswire message. The empirical results
of Chapter 2 provide evidence for an asymmetric reaction of market participants to the arrival of newswire messages of different sentiments. I find higher adverse selection around
negative news messages than around positive news messages. Liquidity increases around
positive and neutral news messages while it has the tendency to decrease around negative

148

5 Conclusion

news messages. Only trading intensity increases around all types of news. An explanation
for the increase in adverse selection around news is that traders aquire costly firm specific
information prior to a news message and that market participants have different capabilities to process new firm specific public information. Both types of behavior lead to higher
information asymmetry among market participants. Liquidity is sustained around positive news as a result of competition for liquidity supply, new positive information is, in
the view of market participants, not disruptive enough for a breakdown of liquidity supply. Ambiguity aversion of a proportion of traders at the TSX is a concept that potentially
explains the asymmetricity of trader behavior with respect to positive and negative news.
The analysis in Chapter 3 is focused on firm specific public information and fragmented
markets, specifically the London Stock Exchange and Chi-X which both offer trading in
FTSE 100 stocks. In such a trading environment, information has multiple opportunities
to translate into prices which yields two research questions. How does firm specific information influence price discovery, liquidity, and trading intensity on individual trading
venues in fragmented markets and how does it influence characteristics of market fragmentation? Again, market participants’ reactions to the daily general firm specific tone
of public information, based on aggregated newswire messages, are asymmetric. For daily
averages, liquidity only decreases on days with predominantely negative firm specific information while it remains stable on positive days. With respect to fragementation characteristics, one result is that overall price discovery shifts to the LSE on positive days in
contrast to neutral days. Also, more trade based information is found to be impounded
into the LSE than on Chi-X on negative days. Within individual order books, results
can be explained with pre-news information gathering of a fraction of market participants
in combination with different post-news information processing capabilities. Consistent
with existing theory (Chowdhry and Nanda, 1991), informed trading, which is higher on
positive and negative news day, gravitates to the LSE, the most liquid market. The empirical analysis also reveals that the market for FTSE 100 stocks is highly liquid and price
discovery is based on relatively efficient processes even on positive and negative news days.
Finally, Chapter 4 takes a more general view on financial markets and considers the association between the firm specific public information flow and stock return comovement,
thus also idiosyncratic volatility. International equity markets all show some amount of
stock return synchronicity which cannot be explained by existing theoretical asset pricing models. Recent research suggests that information production has an influence on
the time-varying properties of stock return comovement (Brockman et al., 2010). Also,

5.2 Outlook

149

comovement varies significantly between different countries. These observations, in combination with the ability to measure a direct proxy for firm specific information through
Thomson Reuters newswire messages, lead to two research questions. How does the relative flow of firm specific information influence stock return comovement and how might
such an association be influenced by country characteristics. Results show that an increase
in the flow of firm specific public information relative to public industry and market information reduces stock return comovement. The firm specific flow of news still includes
information that needs to be capitalized into stock prices and thus increases stocks’ idiosyncratic variability. Additionally, firm specific news might incentivize investors to obtain more firm specific information. I also find that a country’s institutional setting has
an effect on the association between firm specific public information and stock return
comovement. More transparent countries and countries with higher investor protection
show a lower association between firm specific news and comovement. The attenuation of
the association is an indication that in such countries the price already contains more of
the firm specific information found in news.
Newswire messages, such as the Thomson Reuters data used in this thesis, represent
much of the real-time information traders receive. I find that they are a significant source
of information for financial markets. In general, this thesis confirms the important role
that public information has in discovering the efficient price in equity markets and it contributes to the understanding how such public information facilitates efficient financial
markets.

5.2 Outlook
Equity trading has undergone a process of automation and computerization during the
last decades. Now, more than half of all equity trading in developed financial markets is
based on algorithms and computers making buy or sell decisions and placing orders. It is
reasonable to assume that with increasing computing power and available data this computerization will heavily expand into news and information analysis. Already today, traders,
banks, and hedge funds use automatic news analysis to support trading decisions. Recent
news products like the Thomson Reuters Sentiment Engine or News Analytics what it
is called now, Dow Jones Elementized News Feed, and machine readable products from
other information providers directly cater to algorithmic and to high-frequency traders.
Research that would be interesting for regulators and the securities trading industry alike

150

5 Conclusion

could study how an increase in machine driven analysis will change the incorporation of
new information into prices. Whether this trend will increase price efficiency and provide a broader incorporation of information into prices or whether such a development
is a cause for concern has not been answered yet. The question of what happens to financial markets if not only trading but also information based decision making is taken
over by computers, is an interesting area for future research. Another question is whether
an increased linkage of international financial markets through computers will alleviate
differences among financial markets in terms of price discovery and capitalization of information.
This dissertation focuses on equity markets. However, behavior on other markets like
futures, options, foreign exchange, or bond markets might be different and questions arising from this area provide for numerous potential research questions. Some potential explanations for observed trader behavior might also require additional experimental analyses in laboratory settings to control for external influences. For instance, one might gain
more insight into ambiguity aversion in financial markets through controlled economic
experiments but also through datasets that directly identfy individual traders.
This thesis answers some fundamental questions concerning the relation of firm specific
public information and equity markets. But in an ever changing financial market environment, many potential research areas remain and provide for interesting and challenging
research questions in the future.

Appendix A
Sample Firms LSE/Chi-X
Table A.1: Chapter 3 Sample Firms. Table A.1 reports the sample firms for the LSE/Chi-X
analysis including the average daily market capitalization in Million GBP over 2009 and the LSE
and Chi-X Reuters Instrument Codes (RIC).

Firm

LSE RIC

Chi-X RIC

MCap (Mio. GBP)

Anglo American
Associated British Foods
Admiral Group
AMEC
Antofagasta
AstraZeneca
Autonomy Corp
Aviva
BAE Systems
Barclays
British American Tobacco
British Airways
BG Group
British Land Company
BHP Billiton
Bunzl
BP

AAL.L
AALl.CHI
ABF.L
ABFl.CHI
ADML.L ADMLl.CHI
AMEC.L AMECl.CHI
ANTO.L ANTOl.CHI
AZN.L
AZNl.CHI
AUTN.L AUTNl.CHI
AV.L
AVl.CHI
BAES.L
BAESl.CHI
BARC.L BARCl.CHI
BATS.L
BATSl.CHI
BAY.L
BAYl.CHI
BG.L
BGl.CHI
BLND.L BLNDl.CHI
BLT.L
BLTl.CHI
BNZL.L BNZLl.CHI
BP.L
BPl.CHI

23,740.97
5,974.56
2,575.62
2,251.58
6,543.08
38,371.10
3,200.30
9,456.08
12,171.70
26,206.12
36,064.97
1,935.43
35,568.91
3,462.31
94,010.22
1,866.82
96,823.48

continued on next page . . .

A Sample Firms LSE/Chi-X

152

. . . continued from Table A.1
British Sky Broadcasting
BT Group
Cadbury
Carnival
Centrica
Cairn Energy
Cobham
Compass Group
Capita Group
Cable & Wireless
Diageo
Man Group
Eurasion Natural Resources
Experian
G4S
GlaxoSmithKline
Hammerson
Home Retail Group
HSBC Holdings
ICAP
InterContinental Hotels Group
Imperial Tobacco
International Power
Inmarsat
Invensys
Johnson Matthey
Kazakhmys
Kingfisher
Land Securities Group
Legal & General Group
Liberty International
Lloyds Banking Group
Marks and Spencer
Morrison Supermarkets
National Grid
continued on next page . . .

BSY.L
BSYl.CHI 8,714.49
BT.L
BTl.CHI 8,852.04
CBRY.L
CBRYl.CHI 8,527.92
CCL.L
CCLl.CHI 14,242.51
CNA.L
CNAl.CHI 12,596.55
CNE.L
CNEl.CHI 3,306.94
COB.L
COBl.CHI 2,287.52
CPG.L
CPGl.CHI 6,545.79
CPI.L
CPIl.CHI 4,385.28
CW.L
CWl.CHI 3,647.77
DGE.L
DGEl.CHI 22,802.81
EMG.L
EMGl.CHI 4,534.63
ENRC.L ENRCl.CHI 8,679.94
EXPN.L EXPNl.CHI 5,023.78
GFS.L
GFSl.CHI 3,031.86
GSK.L
GSKl.CHI 60,284.53
HMSO.L HMSOl.CHI 2,204.38
HOME.L HOMEl.CHI 2,330.51
HSBA.L
HSBAl.CHI 95,189.54
IAP.L
IAPl.CHI 2,473.58
IHG.L
IHGl.CHI 1,948.98
IMT.L
IMTl.CHI 17,423.58
IPR.L
IPRl.CHI 3,941.14
ISA.L
ISAl.CHI 2,405.31
ISYS.L
ISYSl.CHI 1,871.73
JMAT.L
JMATl.CHI 2,708.75
KAZ.L
KAZl.CHI 4,047.00
KGF.L
KGFl.CHI 4,442.22
LAND.L LANDl.CHI 4,033.45
LGEN.L LGENl.CHI 3,785.92
LII.L
LIIl.CHI 2,233.84
LLOY.L
LLOYl.CHI 19,401.84
MKS.L
MKSl.CHI 5,034.66
MRW.L
MRWl.CHI 6,947.18
NG.L
NGl.CHI 14,540.55

153

. . . continued from Table A.1
NEXT
NXT.L NXTl.CHI
3,104.02
Old Mutual
OML.L OMLl.CHI
4,245.73
Prudential
PRU.L
PRUl.CHI 11,536.79
Pearson
PSON.L PSONl.CHI
5,794.91
Reckitt Benckiser Group
RB.L
RBl.CHI 20,231.46
Royal Bank of Scotland
RBS.L
RBSl.CHI 19,729.10
Royal Dutch Shell A
RDSa.L RDSal.CHI 103,901.99
Royal Dutch Shell B
RDSb.L RDSbl.CHI 103,901.99
Reed Elsevier
REL.L
RELl.CHI
5,574.24
Rexam
REX.L
REXl.CHI
2,081.97
Rio Tinto
RIO.L
RIOl.CHI 48,231.79
Rolls Royce
RR.L
RRl.CHI
7,173.48
Randgold Resources
RRS.L
RRSl.CHI
3,165.69
RSA Insurance Group
RSA.L
RSAl.CHI
4,261.03
SABMiller
SAB.L
SABl.CHI 20,762.99
Sainsbury
SBRY.L SBRYl.CHI
5,815.61
Schroders
SDR.L
SDRl.CHI
2,625.49
Sage Group
SGE.L
SGEl.CHI
2,579.61
Shire
SHP.L
SHPl.CHI
5,448.85
Standard Life
SL.L
SLl.CHI
4,341.68
Smiths Group
SMIN.L SMINl.CHI
3,192.63
Smith & Nephew
SN.L
SNl.CHI
4,460.47
Serco Group
SRP.L
SRPl.CHI
2,153.39
Scottish and Southern Energy
SSE.L
SSEl.CHI 10,370.69
Standard Chartered
STAN.L STANl.CHI 23,873.95
Severn Trent
SVT.L
SVTl.CHI
2,465.39
Thomas Cook Group
TCG.L
TCGl.CHI
1,922.35
Tullow Oil
TLW.L
TLWl.CHI
7,775.12
Tesco
TSCO.L TSCOl.CHI 29,228.68
TUI Travel
TT.L
TTl.CHI
2,721.68
Unilever
ULVR.L ULVRl.CHI 47,908.85
United Utilities Group
UU.L
UUl.CHI
3,379.39
Vedanta Resources
VED.L
VEDl.CHI
4,082.57
Vodafone
VOD.L VODl.CHI 67,698.88
WPP
WPP.L
WPPl.CHI
5,917.92
Xstrata
XTA.L
XTAl.CHI 20,061.58

Appendix B
Sample Data
The following tables report sample data for trade and quote data, depth data, and daily
data as retrieved from the Thomson Reuters DataScope Tick History archive. Table B.1
depicts trade and quote data from the Toronto Stock Exchange. The firm is ‘Research in
Motion’ indicated through the Reuters Instrument Code (RIC) RIM.TO. Bid and ask sizes
are reported in hundreds. The Qualifiers column usually comprises of exchange specific
information. In this sample, ‘Low[USER]’ indicates that this has been the lowest trading
price on that trading day up to that specific point in time. Table B.2 reports depth data for
‘Research in Motion’ traded on the Toronto Stock Exchange. The sample data are from
2006. Still, the thirty data entries only show a snapshot of a little less than three seconds.
Nowadays, the quote update frequency is even higher. A new line in the data appears
as soon as either volume or price changes occur at any depth level reported by the data.
Table B.3 reports Chi-X trade and quote data for ‘Vodafone’ which is listed on the LSE.
Prices are reported in Pence not British Pounds. The raw data are different to the TSX data,
only data fields that change are reported while TSX raw data also includes unchanged data
fields. For instance, if the bid size is updated on Chi-X only the new bid size is reported
while the TSX also features the non changed data fields for bid price, ask size, and ask price
in the same line. Table B.4 reports daily data including traded volume and prices. In this
sample the firm is ‘General Electric’ traded on the New York Stock Exchange. Daily data
are very similar for different exchanges.

Time[G]
14:16:28.018
14:16:28.265
14:16:28.265
14:16:28.683
14:16:28.683
14:16:28.744
14:16:29.225
14:16:29.471
14:16:32.563
14:16:32.563
14:16:32.866
14:16:32.866
14:16:33.560
14:16:33.560
14:16:33.560
14:16:33.560
14:16:33.910
14:16:33.910
14:16:33.910
14:16:33.910
14:16:33.910
14:16:34.153
14:16:34.153
14:16:34.153
14:16:34.274
14:16:35.573
14:16:35.756
14:16:35.756
14:16:35.756
14:16:35.756

Date[G]

01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006

#RIC

RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO

-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4

GMT Offset
Quote
Quote
Quote
Quote
Quote
Quote
Quote
Quote
Quote
Trade
Quote
Quote
Trade
Quote
Trade
Quote
Quote
Quote
Quote
Quote
Quote
Quote
Quote
Quote
Quote
Quote
Quote
Quote
Quote
Quote

Type

200

500
200

71.78
71.75

Volume

71.77

Price

71.75
71.75
71.75
71.75
71.75
71.75
71.75
71.75
71.75
71.75
71.75
71.75
71.75
71.75
71.75

71.75

71.77
71.78

71.79
71.77
71.77
71.77
71.76
71.78
71.77
71.78
71.77

Bid Price

10
8
6
5
4
5
6
7
8
10
8
7
6
5
4

12

8
5

12
6
4
1
1
6
10
5
10

Bid Size

71.77
71.77
71.77
71.77
71.77
71.77
71.77
71.77
71.77
71.77
71.77
71.77
71.77
71.77
71.77

71.78

71.8
71.8

71.8
71.8
71.8
71.8
71.8
71.8
71.8
71.8
71.8

Ask Price

5
5
5
5
5
5
5
5
5
5
5
5
5
5
5

5

2
2

2
2
2
2
2
2
2
2
2

Ask Size

Low[USER]

Low[USER]

Qualifiers

Table B.1: Raw TAQ Data – TSX: This table presents raw trade and quote data from the Toronto Stock Exchange. In this sample the firm is
‘Research in Motion’.

156
B Sample Data

Date[G]

01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006
01-AUG-2006

#RIC

RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO
RIM.TO

13:43:28.013
13:43:28.159
13:43:28.160
13:43:28.161
13:43:28.226
13:43:28.229
13:43:28.229
13:43:28.296
13:43:28.361
13:43:28.361
13:43:28.361
13:43:28.803
13:43:28.912
13:43:28.914
13:43:28.984
13:43:29.126
13:43:29.131
13:43:29.203
13:43:29.355
13:43:29.593
13:43:30.177
13:43:30.180
13:43:30.389
13:43:30.389
13:43:30.606
13:43:30.610
13:43:30.897
13:43:30.901
13:43:30.901
13:43:30.966

Time[G]

-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4
-4

GMT
Offset
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth
Market Depth

Type

73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26
73.26

L1
Bid
Price
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300

L1
Bid
Size
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31
73.31

L1
Ask
Price
900
1000
1000
1000
1100
1100
2200
2600
2600
2600
2600
2600
2600
2600
2600
2600
2600
2600
2600
2600
2600
2600
2600
2600
1500
1500
1500
1500
2600
2600

L1
Ask
Size
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25
73.25

L2
Bid
Price
1000
1000
1000
1000
1000
1000
1000
1000
1000
1000
1100
1100
1100
1100
1100
1100
1100
1000
1000
1000
1000
1000
1000
1000
1000
1000
1100
1100
1100
1100

L2
Bid
Size
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32
73.32

L2
Ask
Price
900
900
900
900
900
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300
300

L2
Ask
Size
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24
73.24

L3
Bid
Price
800
800
800
800
800
800
800
800
800
800
800
800
600
400
200
200
200
200
200
200
200
400
600
600
600
600
600
600
600
600

L3
Bid
Size
73.33
73.33
73.36
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34
73.34

L3
Ask
Price

100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100

L3
Ask
Size

Table B.2: Raw Depth Level 3 Data – TSX: This table presents raw depth data at three levels into the order book from the Toronto Stock
Exchange. Depth data is similar on most exchanges available in the Thomson Reuters DataScope Tick History archive. In this sample the firm
is ‘Research in Motion’.

157

Time[G]
08:47:09.615
08:47:09.922
08:47:13.255
08:47:14.780
08:47:14.780
08:47:14.780
08:47:14.780
08:47:16.306
08:47:16.445
08:47:16.445
08:47:16.445
08:47:16.445
08:47:16.489
08:47:18.464
08:47:21.038
08:47:21.038
08:47:21.038
08:47:21.038
08:47:22.632
08:47:34.921
08:47:34.921
08:47:34.938
08:47:36.143
08:47:37.294
08:47:37.649
08:47:41.579
08:47:41.579
08:47:41.579
08:47:41.579
08:47:41.579

Date[G]

01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009
01-SEP-2009

#RIC

VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI
VODl.CHI

Type
Quote
Quote
Quote
Trade
Quote
Trade
Quote
Quote
Quote
Quote
Quote
Quote
Quote
Quote
Trade
Quote
Quote
Quote
Quote
Trade
Quote
Quote
Quote
Quote
Quote
Trade
Quote
Trade
Quote
Trade

GMT Offset
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
+1
104
1392

14212

8

1413
2000
2000

132.9

132.9

132.9

132.9
132.9
132.9

Volume

132.9

Price

132.9

132.85

132.9

132.85

Bid Price

2000

4000

3413
5413

1413

38593
24493
16419
1421

16419
14212

56793
40593
32519
16419
10419

1392

1496
4505
1496

Bid Size

Ask Price

41357
46357
30157

38257
46357

27148
30157

Ask Size

Qualifiers

Table B.3: Raw TAQ Data – Chi-X: This table presents raw trade and quote data from Chi-X. In this sample the firm is ‘Vodafone’ which is
listed on the London Stock Exchange.

158
B Sample Data

Date[L]
01-AUG-2006
02-AUG-2006
03-AUG-2006
04-AUG-2006
05-AUG-2006
06-AUG-2006
07-AUG-2006
08-AUG-2006
09-AUG-2006
10-AUG-2006
11-AUG-2006
12-AUG-2006
13-AUG-2006
14-AUG-2006
15-AUG-2006
16-AUG-2006
17-AUG-2006
18-AUG-2006
19-AUG-2006
20-AUG-2006
21-AUG-2006
22-AUG-2006
23-AUG-2006
24-AUG-2006
25-AUG-2006
26-AUG-2006
27-AUG-2006
28-AUG-2006
29-AUG-2006
30-AUG-2006

#RIC

GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N
GE.N

Time[L]
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day
End Of Day

Type

No Trades
No Trades

No Trades
No Trades

No Trades
No Trades

No Trades
No Trades

Qualifiers

33.7
33.9
34.25

34
33.9
33.75
33.94
33.72

32.8
33.2
33.35
33.7
33.98

32.68
32.8
32.48
32.35
32.78

32.65
32.55
32.5
32.88

Open

34
34.23
34.44

34.1
34.18
33.86
34.01
33.91

33.43
33.28
33.84
34
34

32.78
32.8
32.72
32.78
32.78

32.67
32.81
32.87
32.99

High

33.7
33.9
34.21

33.77
33.68
33.6
33.77
33.71

32.77
33.04
33.32
33.65
33.8

32.51
32.2
32.24
32.28
32.43

32.48
32.42
32.43
32.6

Low

33.93
34.19
34.27

33.96
33.96
33.79
33.85
33.84

32.82
33.2
33.71
33.92
34

32.69
32.34
32.28
32.67
32.5

32.56
32.6
32.73
32.8

Last

6890900
10029400
6691100

6276500
7653400
9313400
7243000
6236400

12432100
9318700
12016900
9120800
10654900

8162200
11718100
8877000
12836500
8555800

7580000
8597600
9784600
8594900

Volume

Table B.4: Raw Daily Data: This table presents raw daily data. In this sample the firm is ‘General Electric’ traded on the New York Stock
Exchange.

159

Appendix C
RNSE File Format
The following table provides a data description of RNSE data fields. The RNSE data
format includes 41 fields overall, however only the relevant data fields used in this thesis
are introduced here. The following information and description is copied with only minor
changes directly from Thomson Reuters (2008b):

162

C RNSE File Format

TIMESTAMP: The date and time of the news item as timestamped by the NewsScope
Archive, presented in GMT, millisecond precision.
Format: DD MMM YYYY hh:mm:ss.sss
Length: 24
BCAST_REF: Reuters Instrument Code (RIC) of the company for which the scores
apply. Note: While company may trade on a foreign exchange under a different RIC, the
scores are referenced to its Home RIC.
Format: String: <CompanyID>.<Market>
Length: 10
RELEVANCE: A real valued number indicating the relevance of the news item to
the company. It is calculated by comparing how relevant the article is about each of
the companies mentioned in it. For stories with multiple companies mentioned, the
company with the most mentions will have the highest relevance. A company with a
lower amount of mentions will have a lower relevance score.
Format: Real: 0.0-1.0
Length: 10
SENTIMENT: This field indicates the predominant sentiment class for this news item
with respect to this company. The indicated class is the one with the highest probability.
Format: Integer (1: Positive, 0: Neutral, -1: Negative)
Length: 15
LNKD_CNTn: These fields (n: 1-5) contain a list of the number of linguistically
similar items found by the RNSE in each of five history periods: 12 hours, 24 hours,
3 days, 5 days and 7 days. The RNSE takes a “vocabulary fingerprint” of the current
news item and compares it with the fingerprints of other stories from each of the
history periods that mention the current company. The count of linked articles in a
particular time period gives a measure of the novelty of the news being reported – the
higher the linked count value, the less novel the story is. If the count is zero, then the
currrent item can be considered novel as there are no similar items reporting the story
within the history period.
Format: Unsigned Integer
Length: 15
PNAC: Primary News Access Code – a semi-unique story identifier. PNACS are
often reused.
Format: String
Length: 14

References
Akhtar, S., R. Faff, B. Oliver, and A. Subrahmanyam (2011). The power of bad: The negativity bias in Australian consumer sentiment announcements on stock returns. Journal
of Banking and Finance 35(5), 1239–1249.
Andersen, T. G., T. Bollerslev, F. X. Diebold, and C. Vega (2007). Real-time price discovery in global stock, bond and foreign exchange markets. Journal of International
Economics 73(2), 251–277.
Antweiler, W. and M. Z. Frank (2004). Is all that talk just noise? the information content
of internet stock message boards. The Journal of Finance 59(3), 1139–1167.
Bagehot (Pseud.), W. (1971). The only game in town. Financial Analysts Journal 27(2),
12–14.
Barber, B. M. and D. Loeffler (1993). The “dartboard” column: Second-hand information
and price pressure. The Journal of Financial and Quantitative Analysis 28(2), 273–284.
Barber, B. M. and T. Odean (2008). All that glitters: The effect of attention and news on
the buying behavior of individual and institutional investors. The Review of Financial
Studies 21(2), 785–818.
Barclay, M. J. and T. Hendershott (2003). Price discovery and trading after hours. The
Review of Financial Studies 16(4), 1041–1073.
Barclay, M. J., T. Hendershott, and D. McCormick (2003). Competition among trading
venues: Information and trading on electronic communications networks. The Journal
of Finance 58(6), 2637–2665.
Barclay, M. J. and J. B. Warner (1993). Stealth trading and volatility: Which trades move
prices? Journal of Financial Economics 34(3), 281–305.

164

References

Bennett, P. and L. Wei (2006). Market structure, fragmentation, and market quality. Journal of Financial Markets 9(1), 49–78.
Berry, T. D. and K. M. Howe (1994). Public information arrival. The Journal of Finance 49(4), 1331–1346.
Bessembinder, H. (2003a). Issues in assessing trade execution costs. Journal of Financial
Markets 6(3), 233–257.
Bessembinder, H. (2003b). Trade execution costs and market quality after decimalization.
Journal of Financial and Quantitative Analysis 38(4), 747–777.
Bessembinder, H. and H. M. Kaufman (1997). A cross-exchange comparison of execution
costs and information flow for NYSE-listed stocks. Journal of Financial Economics 46(3),
293–319.
Biais, B., P. Hillion, and C. Spatt (1995). An empirical analysis of the limit order book and
the order flow in the paris bourse. The Journal of Finance 50(5), 1655–1689.
Blöhdorn, S. and A. Hotho (2009). Ontologies for machine learning. In S. Staab and
R. Studer (Eds.), Handbook on Ontologies, International Handbooks on Information
Systems, pp. 637–661. Springer Berlin / Heidelberg.
Boehmer, B. and E. Boehmer (2003). Trading your neighbor’s ETFs: Competition or
fragmentation? Journal of Banking and Finance 27(9), 1667–1703.
Bris, A., W. N. Goetzmann, and N. Zhu (2007). Efficiency and the bear: Short sales and
markets around the world. The Journal of Finance 62(3), 1029–1079.
Brockman, P., I. Liebenberg, and M. Schutte (2010). Comovement, information production, and the business cycle. Journal of Financial Economics 97(1), 107–129.
Brown, G. and N. Kapadia (2007). Firm-specific risk and equity market development.
Journal of Financial Economics 84(2), 358–388.
Bushman, R. M., J. D. Piotroski, and A. J. Smith (2004). What determines corporate
transparency? Journal of Accounting Research 42(2), 207–252.

References

165

Cai, C. X., R. Hudson, and K. Keasey (2004). Intra day bid-ask spreads, trading volume
and volatility: Recent empirical evidence from the london stock exchange. Journal of
Business Finance and Accounting 31(5–6), 647–676.
Campbell, J. Y., M. Lettau, B. G. Malkiel, and Y. Xu (2001). Have individual stocks become more volatile? an empirical exploration of idiosyncratic risk. The Journal of Finance 56(1), 1–43.
Cao, C., T. Simin, and J. Zhao (2008). Can growth options explain the trend in idiosyncratic risk? The Review of Financial Studies 21(6), 2599–2633.
Chaboud, A., B. Chiquoine, E. Hjalarsson, and C. Vega (2009). Rise of the machines: Algorithmic trading in the foreign exchange market. FRB International Finance Discussion
Paper No. 980.
Chen, C. W., T. C. Chiang, and M. K. So (2003). Asymmetrical reaction to US stockreturn news: Evidence from major stock markets based on a double-threshold model.
Journal of Economics and Business 55(5–6), 487–502.
Chowdhry, B. and V. Nanda (1991). Multimarket trading and market liquidity. The Review
of Financial Studies 4(3), 483–511.
Chuliá, H., M. Martens, and D. van Dijk (2010). Asymmetric effects of federal funds target
rate changes on S&P100 stock returns, volatilities and correlations. Journal of Banking
and Finance 34(4), 834–839.
Chun, H., J.-W. Kim, J. Lee, and R. Morck (2004). Patterns of comovement: The role of
information technology in the U.S. economy. NBER Working Paper.
De Melo, G. and S. Siersdorfer (2007). Multilingual text classification using ontologies. In
C. Nédellec and C. Rouveirol (Eds.), Advances in Information Retrieval, Volume 4425
of Lecture Notes in Computer Science, pp. 541–548. Amati, Giambattista and Carpineto,
Claudio and Romano, Giovanni.
Debole, F. and F. Sebastiani (2005). An analysis of the relative hardness of Reuters-21578
subsets. Journal of the American Society for Information Science and Technology 56(6),
584–596.

References

166

Degryse, H. (2009). Competition between financial markets in Europe: What can be
expected from MiFID? Financial Markets and Portfolio Management 23(1), 93–103.
Durnev, A., R. Morck, and B. Yeung (2004). Value-enhancing capital budgeting and firmspecific stock return variation. The Journal of Finance 59(1), 65–105.
Durnev, A., R. Morck, B. Yeung, and P. Zarowin (2003). Does greater firm-specific return variation mean more or less informed stock pricing? Journal of Accounting Research 41(5), 797–836.
Ederington, L. H. and J. H. Lee (1993). How markets process information: News releases
and volatility. The Journal of Finance 48(4), 1161–1191.
Ellsberg, D. (1961). Risk, ambiguity, and the savage axioms. The Quarterly Journal of
Economics 75(4), 643–669.
Epstein, L. G. and M. Schneider (2008). Ambiguity, information quality, and asset pricing.
The Journal of Finance 63(1), 197–228.
Epstein, L. G. and M. Schneider (2010). Ambiguity and asset markets. NBER Working
Paper.
Evans, M. D. D. and R. K. Lyons (2008). How is macro news transmitted to exchange
rates. Journal of Financial Economics 88(1), 26–50.
Fang, L. and J. Peress (2009). Media coverage and the cross-section of stock returns. The
Journal of Finance 64(5), 2023–2052.
Fink, J., K. Fink, G. Grullon, and J. Weston (2006). Firm age and fluctuations in idiosyncratic risk. Working Paper.
Fleming, M. J. and E. M. Remolona (1999).

Price formation and liquidity in the

U.S. Treasury market: The response to public information. The Journal of Finance 54(5),
1901–1915.
Foucault, T. and A. J. Menkveld (2008). Competition for order flow and smart order
routing systems. The Journal of Finance 63(1), 119–158.
French, K. R. and R. Roll (1986). Stock return variances: The arrival of information and
the reaction of traders. Journal of Financial Economics 17(1), 5–26.

References

167

Gagliardini, P., P. Porchia, and F. Trojani (2009). Ambiguity aversion and the term structure of interest rates. The Review of Financial Studies 22(10), 4157–4188.
Glosten, L. and P. Milgrom (1985). Bid, ask and transaction prices in a specialist market
with heterogeneously informed traders. Journal of Financial Economics 14(1), 71–100.
Goldstein, M. A. and K. A. Kavajecz (2000). Eights, sixteenths, and market depth: changes
in tick size and liquidity provision on the NYSE. Journal of Financial Economics 56(1),
125–149.
Goldstein, M. A. and K. A. Kavajecz (2004). Trading strategies during circuit breakers and
extreme market movements. Journal of Financial Markets 7(3), 301–333.
Goldstein, M. A., A. V. Shkilko, B. F. Van Ness, and R. A. Van Ness (2008). Competition
in the market for NASDAQ securities. Journal of Financial Markets 11(2), 113–143.
Green, T. C. (2004). Economic news and the impact of trading on bond prices. The Journal
of Finance 59(3), 1201–1233.
Groß-Klußmann, A. and N. Hautsch (2011). When machines read the news: Using automated text analytics to quantify high frequency news-impled market reactions. Journal
of Empirical Finance 18(2), 321–340.
Guo, H. and R. Savickas (2008). Average idiosyncratic volatility in G7 countries. The
Review of Financial Studies 21(3), 1259–1296.
Hamao, Y., J. Mei, and Y. Xu (2003). Idiosyncratic risk and the creative destruction in
japan. NBER Working Paper.
Hameed, A., R. Morck, J. Shen, and B. Yeung (2010). Information, analysts, and stock
return comovement. NBER Working Paper.
Hansen, P. and A. Lunde (2005). A realized variance for the whole day based on intermittent high-frequency data. Journal of Financial Econometrics 3(4), 525–554.
Harris, L. E. (1994). Minimum price variations, discrete bid-ask spreads, and quotation
sizes. The Review of Financial Studies 7(1), 149–178.
Harris, M. and A. Raviv (1993). Differences of opinion make a horse race. The Review of
Financial Studies 6(3), 473–506.

168

References

Hasbrouck, J. (1991a). Measuring the information content of stock trades. The Journal of
Finance 46(1), 179–207.
Hasbrouck, J. (1991b). The summary informativeness of stock trades: An econometric
analysis. The Review of Financial Studies 4(3), 571–595.
Hasbrouck, J. (1995). One security, many markets: Determining the contributions to
price discovery. The Journal of Finance 50(4), 1175–1199.
Höchstötter, M., R. Riordan, and A. Storkenmaier (2011). International stock market
comovement and news. Working Paper.
Hendershott, T., C. M. Jones, and A. J. Menkveld (2011). Does algorithmic trading improve liquidity? The Journal of Finance. forthcoming.
Hendershott, T. and R. Riordan (2009). Algorithmic trading and information. Working
Paper.
Hess, D., A. Kempf, and M. Malinowska (2008). Liquidity provision in periods of high
information flow. Working Paper.
Huang, R. D. (2002). The quality of ecn and nasdaq market maker quotes. The Journal of
Finance 57(3), 1285–1319.
Hutton, A. P., A. J. Marcus, and H. Tehranian (2009). Opaque financial reports, R2 , and
crash risk. Journal of Financial Economics 94(1), 67–86.
International Telecommunication Union (2009). Measuring the information society - the
ICT development index. White Paper.
Irvine, P. J. and J. Pontiff (2009). Idiosyncratic return volatility, cash flows, and product
market competition. The Review of Financial Studies 22(3), 1149–1177.
Jin, L. and S. C. Myers (2006). R2 around the world: New theory and new tests. Journal
of Financial Economics 79(2), 257–292.
Joachims, T. (1998). Text categorization with support vector machines: Learning with
many relevant features. In C. Nédellec and C. Rouveirol (Eds.), Machine Learning:
ECML-98, Volume 1398 of Lecture Notes in Computer Science, pp. 137–142. Springer
Berlin / Heidelberg.

References

169

Jones, C. M. and M. L. Lipson (2001). Sixteenths: direct evidence on institutional execution costs. Journal of Financial Economics 59(2), 253–278.
Kandel, E. and N. D. Pearson (1995). Differential interpretation of public signals and trade
in speculative markets. Journal of Political Economy 103(4), 831–872.
Karolyi, G. A., K.-H. Lee, and M. A. van Dijk (2009). Commonality in returns, liquidity,
and turnover around the world. Working Paper.
Khandaker, S. and R. Heaney (2009). Do emerging markets have higher stock synchronicity? the international evidence. Journal of Business and Policy Research 4(1), 79–98.
Kim, O. and R. E. Verrecchia (1991). Market reactions to anticipated announcements.
Journal of Financial Economics 30(2), 273–309.
Kim, O. and R. E. Verrecchia (1994). Market liquidity and volume around earnings announcements. Journal of Accounting and Economics 17(1–2), 41–67.
Klar, J. and I. van den Bongard (2008). Determinants of the bid-ask spread and the role of
designated sponsors: Evidence for Xetra. Working Paper.
Krinsky, I. and J. Lee (1996). Earnings announcements and the components of the bid-ask
spread. The Journal of Finance 51(4), 1523–1535.
Kyle, A. S. (1985). Continuous auctions and insider trading. Econometrica 53(6), 1315–
1335.
La Porta, R., F. Lopez-de Silanes, and A. Shleifer (1998). Law and finance. Journal of
Political Economy 106(6), 1113–1155.
La Porta, R., F. Lopez-de-Silanes, A. Shleifer, and R. Vishny (2000). Investor protection
and corporate governance. Journal of Financial Economics 58(1–2), 3–27.
Lee, C. and M. Ready (1991). Inferring trade direction from intraday data. The Journal of
Finance 46(2), 733–746.
Leippold, M., F. Trojani, and P. Vanini (2008). Differences of opinion make a horse race.
The Review of Financial Studies 21(6), 2567–2597.

170

References

Leopold, E. and J. Kindermann (2002). Text categorization with support vector machines.
how to represent texts in input space? Machine Learning 46(1), 423–444.
Lewis, D. D., Y. Yang, T. G. Rose, and F. Li (2004). RCV1: A new benchmark collection
for text categorization research. Journal of Machine Learning Research 5, 361–397.
Li, K., R. Morck, F. Yang, and B. Yeung (2004). Firm-specific variation and openness in
emerging markets. The Review of Economics and Statistics 86(3), 658–669.
Liu, P., S. D. Smith, and A. A. Syed (1990). Stock price reactions to the wall street journal’s
securities recommendations. Journal of Financial and Quantitative Analysis 25(3), 399–
410.
Madhavan, A. (2000). Market microstructure: A survey. Journal of Financial Markets 3(3),
205–258.
Madhavan, A., M. Richardson, and M. Roomans (1997). Why do security prices change? a
transaction-level analysis of NYSE stocks. The Review of Financial Studies 10(4), 1035–
1064.
Mendelson, H. (1987). Consolidation, fragmentation, and market performance. Journal of
Financial and Quantitative Analysis 22(2), 189–207.
Mitchell, M. L. and J. H. Mulherin (1994). The impact of public information on the stock
market. The Journal of Finance 49(3), 923–950.
Morck, R., B. Yeung, and W. Yu (2000). The information content of stock markets: Why
do emerging markets have synchronous stock price movements. Journal of Financial
Economics 58(1–2), 215–260.
Morse, D. (1981). Price and trading volume reaction surrounding earnings announcements: A closer examination. Journal of Accounting Research 19(2), 374–383.
Newey, W. K. and K. D. West (1987). A simple, positive semi-definite, heteroskedasticity
and autocorrelation consistent covariance matrix. Econometrica 55(3), 703–708.
Niederhoffer, V. (1971). The analysis of world events and stock prices. The Journal of
Business 44(2), 193–219.

References

171

Niessen, A. (2007). Media coverage and macroeconomic information processing. SFB 649
Discussion Paper 2007-011.
Pagano, M. (1989).

Trading volume and asset liquidity.

Quarterly Journal of Eco-

nomics 104(2), 255–274.
Patell, J. M. and M. A. Wolfson (1982). Good news, bad news, and the intraday timing of
corporate disclosures. The Accounting Review 57(3), 509–527.
Petersen, M. A. (2009). Estimating standard errors in finance panel data sets: Comparing
approaches. The Review of Financial Studies 22(1), 435–480.
Piotroski, J. D. and B. T. Roulstone (2004). The influence of analysts, institutional investors, and insiders on the incorporation of market, industry, and firm-specific information into stock prices. The Accounting Review 79(4), 1119–1151.
Ranaldo, A. (2006). Intraday market dynamics around public information arrivals. Swiss
National Bank Working Paper.
Read, D. (1999). The Power of News: The History of Reuters (2 ed.). Oxford: Oxford
University Press.
Riordan, R., A. Storkenmaier, and M. Wagener (2010a). Fragmentation, competition and
market quality: A post-mifid analysis. Working Paper.
Riordan, R., A. Storkenmaier, and M. Wagener (2010b). Public information arrival: Price
discovery and liquidity in electronic limit order markets. Working Paper.
Roll, R. (1988). R2 . The Journal of Finance 43(3), 541–566.
Ronis, D. L. and E. R. Lipinski (1985). Value and uncertainty as weighting factors in
impression formation. Journal of Experimental Social Psychology 21, 47–60.
Ryan, P. and R. J. Taffler (2007). Are economically significant stock returns and trading
volumes driven by firm-specific news releases. Journal of Business Finance & Accounting 31(1), 49–82.
Soroka, S. N. (2006). Good news and bad news: Asymmetric responses to economic
information. The Journal of Politics 68(2), 372–385.

172

References

Stock, J. H. and M. W. Watson (1988). Testing for common trends. Journal of the American
Statistical Association 83(404), 1097–1107.
Storkenmaier, A., R. Riordan, C. Weinhardt, and R. Studer (2010). The impact of economic news on information and liquidity in electronic futures trading. In T. Dreier,
J. Krämer, R. Studer, and C. Weinhardt (Eds.), Information Management and Market
Engineering: Vol. II. Studies on eOrganisation and Market Engineering, pp. 37–54. KIT
Scientific Publishing.
Storkenmaier, A., M. Wagener, and C. Weinhardt (2010). Public information in fragmented markets. Working Paper.
Tetlock, P. C. (2007). Giving content to investor sentiment: The role of media in the stock
market. The Journal of Finance 62(3), 1139–1167.
Tetlock, P. C. (2008). All the news that’s fit to reprint: Do investors react to stale information. Working Paper.
Tetlock, P. C., S. A. Macskassy, and M. Saar-Tsechansky (2008). More than words: Quantifying language to measure firms’ fundamentals. The Journal of Finance 63(3), 1427–1467.
Thompson, R. B., C. Olson, and R. Dietrich (1987). Attributes of news about firms:
An analysis of firm-specific news reported in the wall street journal index. Journal of
Accounting Research 25(2), 245–274.
Thompson, S. B. (2011). Simple formulas for standard errors that cluster by both firm and
time. Journal of Financial Economics 99(1), 1–10.
Thomson Reuters (2008a). Reuters Newsscope Sentiment Engine: Guide to sample data
and system overview. Thomson Reuters White Paper.
Thomson Reuters (2008b). Reuters Newsscope Sentiment Engine: Output image and file
format. Thomson Reuters White Paper.
Tong, S. and D. Koller (2001). Support vector machine active learning with applications
to text classification. Journal of Machine Learning Research 2, 45–66.
Treynor, J. (1995). The only game in town (reprint). Financial Analysts Journal 51(1),
81–83.

References

173

Veldkamp, L. L. (2005). Slow boom, sudden crash. Journal of Economic Theory 124(2),
230–257.
Veldkamp, L. L. (2006). Information markets and the comovement of asset prices. Review
of Economic Studies 73(3), 823–845.
Veronesi, P. (1999). Stock market overreaction to bad news in good times: A rational
expectations equilibrium model. The Review of Financial Studies 12(5), 975–1007.
White, H. (1980). A heteroskedasticity-consistent covariance matrix estimator and a direct
test for heteroskedasticity. Econometrica 48(4), 817–838.
Willard, K. L., T. W. Guinnane, and H. S. Rosen (1996). Turning points in the civil war:
Views from the greenback market. The American Economic Review 86(4), 1001–1018.
Wurgler, J. (2000). Financial markets and the allocation of capital. Journal of Financial
Economics 58(1–2), 187–214.

The last decades have seen dramatic changes in trading technology and
the way that financial markets operate. As trading technology advances,
news providers have kept pace and deliver news to market participants
around the world within fractions of a second using electronic systems.
Currently, most news is still interpreted by humans but news providers
have started to offer newswire products with machine learning systems
that specifically cater to algorithmic traders. In practice, newswire messages make up a major part of the public information set available to
investors. This book studies how newswire messages impact modern electronic equity markets.

ISBN 978-3-86644-694-6

ISBN 978-3-86644-694-6

9 783866 446946