News sentiment and sovereign credit risk

• Author: Matthias Uhl,Yining Shi,Nina M. Gotthelf,Lara Cathcart
• Published date: 01 March 2020
• DOI: http://doi.org/10.1111/eufm.12219
• Date: 01 March 2020

Eur Financ Manag. 2020;26:261–287. wileyonlinelibrary.com/journal/eufm © 2019 John Wiley & Sons Ltd.

|

261

DOI: 10.1111/eufm.12219

ORIGINAL ARTICLE

News sentiment and sovereign credit risk

Lara Cathcart

1

|

Nina M. Gotthelf

2

|

Matthias Uhl

3

|

Yining Shi

4

1

Finance Department, Imperial College

Business School, South Kensington

Campus, SW7 2AZ, London

2

Imperial College London, University of

Zürich, Switzerland

3

University of Zürich and UBS Asset

Management, Switzerland

4

School of Banking & Finance, University

of International Business and Economics,

10 Hui Xin East Road, Beijing, 100029,

China

Abstract

We explore the impact of media content on sovereign

credit risk. Our measure of media tone is extracted from

the Thomson Reuters News Analytics database. As a

proxy for sovereign credit risk we consider credit default

swap (CDS) spreads, which are decomposed into their

risk premium and default risk components. We find that

media tone explains and predicts CDS returns and is a

mixture of noise and information. Its effect on risk

premium induces a temporary change in investors’

appetite for credit risk exposure, whereas its impact on

the default component leads to reassessments of the

fundamentals of sovereign economies.

KEYWORDS

CDS, credit risk premium, media tone, sovereign risk

JEL CLASSIFICATION

G12; G15

1

|

INTRODUCTION

Does news sentiment impact sovereign credit risk? Does it improve our understanding of

countries’fundamentals? Does it proxy for investor sentiment? We investigate these questions

by using an extensive sovereign credit default swap (CDS) data set and the ratings of news

content (i.e. sentiment) from Thomson Reuters News Analytics (TRNA). We consider sovereign

CDS contracts as a proxy for country‐specific credit risk. These represent an insurance contract

against sovereign default or restructuring events and are generally more liquid than the

underlying bond. In the TRNA database news items are rated in terms of sentiment (positive or

negative) in real time using a highly sophisticated neural network which provides an

EUROPEAN

FINANCIAL MANAGEMENT

We thank John Doukas (the Editor) and two anonymous referees Darrell Duffie, Lasse Pedersen, seminar participants

at the University of Zürich, and the World Finance & Banking Symposium 2016 in Dubai for valuable input and

suggestions. All errors remain our own.

improvement over traditional approaches (such as bag‐of‐words). Furthermore, TRNA reflects a

more accurate representation of the news set used by actual investors, as it is a commercial

product that is sold directly to subscribers.

Numerous studies have explored sovereign credit risk and several of its determinants.

1

In

particular, Longstaff et al. (2011) highlight the high level of commonality in sovereign credit

spreads and find that they are mainly explained by global factors.

2

Country‐specific

fundamentals have less explanatory power. Coupled with global factors, behavioral measures,

such as investor sentiment, have also been found to influence sovereign credit risk.

3

Nevertheless, it is crucial to point out that we focus on news sentiment instead of investor

sentiment as in the aforementioned papers. The news sentiment is based on textual analysis of

news events with pre‐defined and trained linguistic sentiment combinations or expert

consensus. Thus, the news sentiment in our paper can be treated as more of the exogenous

information nature of news and is ex ante to the market reaction. On the other hand, investor

sentiments are market‐based proxies of sentiment that already contain investors’perceptions

and even reactions toward the reported news or articles. Our work contributes to the findings in

Apergis, Lau, and Yarovaya (2016) that news sentiment affects sovereign credit risk. We differ

from their paper in the sense that we focus on 25 developed and less developed countries in

various geographical regions from 2003 to 2014, while their paper studies only five European

sovereign nations from 2009 to 2012.

4

Furthermore, the news sentiment in our paper is rather

comprehensive and advanced, from a professional data vendor. They use their own calculated

news sentiment.

The ability of media content to impact equity markets has recently received considerable

attention in the literature. In particular, Tetlock (2007) examines how qualitative information is

incorporated into aggregate market valuations, and Garcia (2013) shows that the predictability

of stock returns using news content is concentrated in recessions. Dougal, Engelberg, Garcia,

and Christopher (2012) identify a causal relationship between financial reporting and stock

market performance. Engelberg and Parsons (2011) find that local media coverage strongly

predicts local trading, and that local trading is strongly related to the timing of local reporting.

Uhl, Pedersen, and Malitius (2015) find a longer‐run effect of news sentiment on equities with

weekly data. Tetlock, Saar‐Tsechansky, and Macskassy (2008) examine the impact of negative

words on individual S&P 500 firms, Ahmad, Han, Hutson, Kearney, and Liu (2016) study the

relation between firm‐specific media tone and firm‐level return using 20 large US firms, and

Hillert, Jacobs, and Müller (2014) find that firms particularly covered by the media exhibit

stronger momentum and that this effect depends on media tone. Hence, media coverage and

sentiment in news influence investors and potentially exacerbate investor biases.

To investigate the impact of media tone in the sovereign CDS market, we first construct a

global “news sentiment”variable from TRNA by filtering news according to global debt

markets, and regional classifiers, such as Europe, Latin America and Asia. The motivation for a

global “news sentiment”variable is built on results established in the literature which are

suggestive of increasing economic integration across countries, growing dependence on global

1

These include, among others, Edwards (1984), (1986), Berg and Sachs (1988), Boehmer and Megginson (1990), Duffie, Pedersen, and Singleton (2003),

Longstaff, Pan, Pedersen, and Singleton (2011), Pan and Singleton (2008), Remolona, Scatigna, and Wu (2007), Jeanneret (2015), Badaoui, Cathcart, and El‐

Jahel (2013) and Monfort and Renne (2014).

2

The most significant variables for CDS spreads have been found to be the US stock market, high‐yield spreads and the VIX index.

3

See Georgoutsos and Migiakis (2013), Tang and Yan (2013), Aizenman, Jinjarak, Lee, and Park (2016) and Lee, Kim, and Park (2016).

4

For the classification of developed and less developed countries see Unted Nations (2018).

262

|

EUROPEAN

FINANCIAL MANAGEMENT

CATHCART ET AL.