Public News Arrival and Cross‐Asset Correlation Breakdown

• Date: 01 September 2018
• Author: Jing Yu,Wai‐Man Liu,Kin‐Yip Ho
• Published date: 01 September 2018
• DOI: http://doi.org/10.1111/irfi.12156

Public News Arrival and Cross-Asset

Correlation Breakdown*

KIN-YIP HO

†

,WAI-MAN LIU

†

AND JING YU

‡

†

Research School of Finance, Actuarial Studies and Statistics, The Australian National

University (ANU), Canberra, Australia and

‡

Discipline of Accounting and Finance, The University of Western Australia Business

School, Perth, Australia

ABSTRACT

This study models and tests empirically the role of public news arrivals in

the quote matching across single-stock futures and underlying stock

markets—a trading strategy often adopted by algorithmic traders. Our model

suggests that quote return correlation across these two markets breaks down

when the news uncertainty is sufficiently large and futures market makers

switch from automating the quote matching process to manually analyze,

monitor, and update quotes. We show empirically that the breakdown is

more prominent for large stocks, and this effect of firm size falls during peri-

ods of high-market volatility. Our empirical results are robust to the effect of

distraction due to extraneous news events.

JEL Codes: G10; G12; G14

Accepted: 17 August 2017

I. INTRODUCTION

High speed of trading and an increasing pace of information dissemination in

the marketplace are two most significant developments of our financial envi-

ronment largely due to the rapid rise of algorithmic trading.

1

As an emerging

* The authors would like to thank Sean Anthonisz, Peter Bossaerts, Jonathan Brogaard, Kyoung-

Shik Eom, Carole Comerton-Forde, Doug Foster, Bruce Grundy, Richard Heaney, Terrence Hender-

shott, Ivan Indriawan, Philip Maymin, Phong Ngo, Geoff Warren, Jaeho Yoon, Yinggang Zhou, and

conference participants at 2015 Auckland Finance Meeting, the 10th International Paris Finance

Meeting, 7th Annual Conference on Asia-Pacific Financial Markets of the Korean Securities Associa-

tion, and seminar participants at The Australian National University (ANU), Chinese University of

Hong Kong, Deakin University, La Trobe University, Ewha Woman’s University, and University of

Sydney. We would like to thank Yanlin Shi from ANU for excellent research assistance.

1 By 2009, algorithmic trading has dominated a substantial portion of equity trades in the U.S., with

more than 70% of all equity trades being generated by orders from computer models reacting in

sub-seconds to market news and real-time price movements (Donefer 2010; Hendershott

et al. 2011). These strategies are estimated to represent industry revenues of around US$7–$9 bil-

lion. For the first quarter of 2010, at least one third of the order book executions on the London

Stock Exchange (LSE) are resulted from algorithmic trading. The market share of algorithmic

© 2017 International Review of Finance Ltd. 2017

International Review of Finance, 18:3, 2018: pp. 411–451

DOI: 10.1111/irfi.12156

form of trading scheme, algorithmic trading utilizes computer-based algorithms

to implement high-frequency trading strategies without human intervention. A

young but growing literature emerges to understand the market impact of high-

frequency trading environment. Despite their mixed conclusions, an important

assumption underlying the theoretical work in this strand of literature is that

algorithmic trading can respond promptly to market news events and process

news faster than their human counterparts (e.g., Biais et al. 2011; Foucault

et al. 2013; Jovanovic and Menkveld 2012).

However, the market reactions to the erroneous report on the United Airlines

(UAL) prompt us to pause and rethink about this assumption: on September

8, 2008, a headline announcing the bankruptcy of the airline hit the news feeds

by mistake. Within a short span of 12 min, the UAL share price plummeted by

75% to $3 a share before it subsequently recovered. As suggested by the

New York Times on September 13, 2008, “human error seems to have played

only a minor role. The financial damage was mostly the result of the interplay

between the algorithms that search and compile information from the Web

and the ones that Wall Street firms and hedge funds use to make trades auto-

matically.”This story challenges our conventional view of how algorithmic

trading handles information flows in reality. The immediate questions are: can

algorithmic trading conducted by machines react appropriately to the arrivals

of complex public news which require advanced analytical interpretation?

Under what circumstances are these news arrivals relevant?

Our study attempts to tackle these questions by modeling and testing empiri-

cally the role of public news arrivals in a trading environment populated by

algorithmic trading. Broadly speaking, there are two types of algorithmic trad-

ing. One type of algorithmic traders uses computers to reduce implicit transac-

tions costs (e.g., Cvitanic and Kirilenko 2010) while the other plays the role of

arbitrageurs who constantly seek for mispricing opportunities (e.g., Jarrow and

Protter 2012). Our paper relates to the latter type. Specifically, we are interested

in studying the trade-offs that market makers of two closely related assets face

when they use algorithmic quote updating to reduce mispricing risk, and fac-

tors that might alter the relationship between news arrivals and the return cor-

relations between these two assets. The asset pair used in our empirical setting

is the single-stock futures (SSF) contracts and their underlying stocks since in

normal circumstances their return correlation should be close to perfect and

they do not vary over time. Furthermore, unlike stock options, the cost-of-carry

for stock futures is fairly predictable, which gives us a cleaner measure of corre-

lation. Also, studying the SSF market enables us to examine how cross-sectional

firm characteristics affect the news-correlation dynamics.

trading on LSE is obtained from the exchange’s responses to Committee of European Securities

Regulators’Call for evidence on micro-structural issues of the European equity markets (April

30, 2010). Website: http://ww w.londons tockexcha nge.com/a bout-the-e xchange/ regulatory /

responsetocesrscallforevidenceonmicro-structuralissuesoftheeuropeanequitymarkets.pdf

(Viewed July 4, 2013). Similarfast growing pattern of algorithmic tradingis also seen in foreign

exchange and financialderivatives markets (Gomber et al. 2011; Chaboud et al.2014).

© 2017 International Review of Finance Ltd. 2017412

International Review of Finance

The intuition behind our model is straightforward. By the cost-of-carry rela-

tion, futures market makers peg the quotes to the quotes of the underlying

automatically using computerized algorithms. However, futures market makers

may shift from quote pegging to monitoring and analyzing the content of the

news feed upon the arrival of public news if the news content is vague about

the asset’s fundamental value, and may fine tune the parameters of the algo to

reflect the news content. As an anecdotal support for this intuition underpin-

ning our model, we note that Interactive Brokers carries out trades of multiple

assets including SSF using a trading program known as Trade Workstation

(TWS). To trade based on the news arrivals, TWS allows the trader to define

algorithmic conditions in response to news events.

2

Studying the news feed

and analyzing its impact on prices can be a costly exercise because the correct

interpretation of an announcement requires human attention and processing

(see Foucault et al. 2003; Liu 2009 for theory; see Chakrabarty and Moulton

2012 for empirical support). Because news monitoring and analysis is costly,

our model suggests that, futures market makers will widen the spread to com-

pensate for its cost, and ultimately cause a momentary breakdown in spot-

futures return correlation.

3

Our theoretical model yields three empirical implications. First, the strong

contemporaneous return correlation between spot and futures declines as pub-

lic news arrives and the dispersion of agreements on public news heightens,

provided that the cost of news analysis is not too high. Second, the return cor-

relation breakdown is more likely to occur for large stocks which are character-

ized by lower costs of monitoring (Size Effect). Finally, we show that the Size

Effect weakens when the market participants are distracted (i.e., when the

opportunity costs of not monitoring other stocks are high).

We test these empirical implications using high-frequency transactions data

of all US SSF contracts listed on One-Chicago Futures Exchange and their

underlying stocks from Thomson Reuters Tick History (TRTH) database, and

real-time public news database from RavenPack News Analytics Dow Jones Edi-

tion (RavenPack) database between July 20, 2009 (when the SSF quote data first

becomes available) and March 19, 2010 (when the March 2010 contract

expires). To gauge the realized spot-futures return correlations, we design two

conditional correlation measures that are computed with a forward rolling win-

dow of six 10-min time intervals. The first measure is based on the logarithmic

returns of spot and futures quotes and the second measure is based on the resi-

duals from filtering the logarithmic returns of spot and futures quotes through

the VEC-BEKK model. The latter construct is intended to address the cost of

carry relation and the cointegrating properties of the spot and futures prices.

2 Further details are available on the TWS Guide at http://www.interactivebrokers.com.

3 One might argue that at the arrival of news, market makers do not shut down their algo-

rithms entirely, but instead use some types of news analytics or other forms of black box algo-

rithms to avoid human intervention. This may be true but often it is necessary for market

makers to reset parameters of their algorithms to reflect the news contents, and this requires,

albeit reduced, human intervention.

© 2017 International Review of Finance Ltd. 2017 413

News and Cross-Asset Correlation Breakdown