Media Attention vs. Sentiment as Drivers of Conditional Volatility Predictions: An Application to Brexit

doi:10.1016/j.frl.2021.101943

Finance Research Letters

Volume 42, October 2021, 101943

https://doi.org/10.1016/j.frl.2021.101943 Get rights and content

Highlights

•
We use data on on-line media coverage and tone of the Brexit referendum
•
We test whether media coverage and/or tone provide forecasting improvements inthe prediction of the conditional variance of weekly FTSE 100 returns
•
Versions of symmetric and asymmetric GARCH models augmented to include mediacoverage and especially media tone outperform traditional GARCH

Abstract

Using data on international, on-line media coverage and tone of the Brexit referendum, we test whether it is media coverage or tone to provide the largest forecasting performance improvements in the prediction of the conditional variance of weekly FTSE 100 stock returns. We find that versions of standard symmetric and asymmetric Generalized Autoregressive Conditional Heteroskedasticity (GARCH) models augmented to include media coverage and especially media tone scores outperform traditional GARCH models both in- and out-of-sample.

Introduction

It is well understood that media coverage (a proxy for investors’ attention) and tone (a proxy for market sentiment) are important drivers of trading volumes, stock returns, and to some extent, also risk (see, among others, Broadstock and Zhang, 2019; Brown and Cliff, 2005; Da, Engelberg, and Gao, 2011; Tetlock, 2007). While the evidence on the predictive power of coverage and sentiment for stock returns and trading volume is pervasive, the available results concerning the connection between media attention and tone and standard measures of risk, such as conditional variance, were initially tenuous at best. For instance, Mitchell and Mulherin (1994) and Berry and Howe (1994) relate aggregate stock market volume and volatility to broad measures of news about firms and the economy. Both studies reported weak correlations of less than 0.12 between market volatility and the news items. However, Antweiler and Frank (2004) studied the frequency and tone of stock message-board posts on Yahoo! Finance and Raging Bull for 45 large US stocks in the year 2000. They find that message-board posting frequency predicts return volatility and trading volume, even controlling for the frequency of WSJ stories.

We perform analyses that to the best of our knowledge are new to test whether on-line media attention and sentiment have any predictive power for risk, measured as the conditional variance of returns of the FTSE 100 stock index (and, as a robustness check, of the FTSE 250 stock index). We base our investigation on the number of articles and the tone extracted from more than half a million on-line news items concerning the June 2016 Brexit referendum vote over a daily 2015-2018 sample, available from Global Database of Events, Language, and Tone (GDELT). Notably, we conduct our analysis within a classical Generalized Autoregressive Conditional Heteroskedasticity (GARCH) framework, i.e., we seek to measure whether and how the in- and out-sample predictive performance of off-the shelf GARCH can be improved by augmenting the model to include exogenous scores that reflect media coverage and tone.

Importantly, there is some theoretical backing for our empirical tests. For instance, Andrei and Hasler (2014) develop an asset-pricing model that derives a link between asset prices, as well as volatility, and investors' attention. In their model, the level of attention varies with economic conditions, and the model shows that higher attention and higher uncertainty are associated with higher stock return variance. This link between attention and volatility is supported by some empirical literature (see, e.g. Da, Engelberg, and Gao, 2011; Dimpfl and Jank, 2016; Guidolin et al., 2017). Paradoxically, we know less about the link between sentiment and equity risk, even though Da, Engelberg, and Gao (2015) construct a new measure of investor sentiment (the Financial and Economic Attitudes Revealed by Search, FEARS) index, by aggregating the volume of Google Trends queries related to concerns such as recession, unemployment and bankruptcy and report that FEARS predicts volatility. ¹

In the perspective of showing the importance of attention and sentiment in forecasting the volatility of asset returns, the Brexit vote represents a highly relevant case for several reasons. Indeed, Brexit has undoubtfully raised concerns among investors (see Adesina, 2017), while Arshada, Rizvib, and Haroon (2020) find that the efficiency of the British stock market worsened drastically because of the uncertainty caused by the Brexit vote. Our key result is that while recent findings that media attention increases the conditional variance of asset returns are confirmed, it is mostly the (absolute value of the) tone of the on-line media flow that generates risk.² Therefore, it is not just the presence of media chatter and coverage that generates risk, but also (or especially) the fact the news come to reflect strong views, with a specific tone that seems to affect stock prices more, thus offering precise forecasting power.

Section 2 describes our data and the alternative models that we propose to forecast conditional volatility. Section 3 discusses the empirical results. Section 4 concludes.

Section snippets

Data

Our source of data is the Global Database of Events, Language, and Tone (GDELT), which collects news from print, broadcast, and online media in more than 100 languages and from all over the world and analyze them using natural language processing algorithms.³ To construct the series employed in this paper we have relied on the Global Knowledge Graph tool (henceforth, GKG). The GKG is a large collection of press articles organized around a “nameset”, a set of

Methodology

The first specification that we propose is a GARCHX obtained by adding each of the two news-related series as exogenous variables to a standard GARCH: $R_{t + 1} = σ_{t + 1 | t} z_{t + 1} z_{t + 1} \sim I I D N (0, 1)$ $σ_{t + 1 | t}^{2} = ω + α R_{t}^{2} + β σ_{t | t - 1}^{2} + γ x_{i, t}^{2} i = 1, 2$ where R_t is the (daily) log return defined as $R_{t} \equiv \ln (\frac{P_{t}}{P_{t - 1}}), P_{t}$ is the value of the FTSE 100 index at time t, z_t is a normally distributed random shock, and $σ_{t | t - 1}^{2}$ is the conditional variance at time t given the information set at time t − 1, x_{i = 1, t} = tone_t while $x_{i = 2, t} = A r t_g r o w t h$

In-sample forecasting results

Table 2 shows the in-sample estimates and the measures of in-sample fit for the models discussed in Section 3. Notably, the models including the growth in the number of articles either in a symmetric or an asymmetric fashion fail to outperform a standard GARCH(1,1) model, in-sample.⁹

Conclusion

We propose alternative augmented GARCH models that include measures of the coverage and the tone of the media for a specific event, the 2016 Brexit referendum. We assess the predictions of these models vs. a standard GARCH. The models that include the media tone outperform GARCH in three specifications: when it is always considered (GARCHX), when it is considered above q threshold of previous predicted variance (GARCHND), and under an asymmetric effect related to the sign of the exogenous

Author's Statement (on Behalf of All Authors)

1. The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

2. All authors have contributed in equal shares to all the parts of this submission.

3. The data have been duly shared and made available.

References (21)

T. Adesina
Estimating volatility persistence under a Brexit-vote structural break
Finance Research Letters
(2017)
D.C. Broadstock et al.
Social-media and intraday stock returns: The pricing power of sentiment
Finance Research Letters
(2019)
J.Y. Campbell et al.
No news is good news: An asymmetric model of changing volatility in stock returns
Journal of Financial Economics
(1992)
R. Davies et al.
The heterogeneous impact of Brexit: Early indications from the FTSE
European Economic Review
(2018)
C. Kearney et al.
Textual sentiment in finance: A survey of methods and models
International Review of Financial Analysis
(2014)
D. Shen et al.
Does twitter predict Bitcoin?
Economics Letters
(2019)
Rizvi Arshad et al.
Impact of Brexit vote on the London stock exchange: A sectorial analysis of its volatility and efficiency
Finance Research Letters
(2020)
D. Andrei et al.
Investor attention and stock market volatility
Review of Financial Studies
(2014)
W. Antweiler et al.
Is all that talk just noise? The information content of Internet stock message boards
Journal of Finance
(2004)
A. Belke et al.
Policy uncertainty and international financial markets: the case of Brexit
Applied Economics
(2018)

There are more references available in the full text version of this article.

Cited by (7)

Google search trends and stock markets: Sentiment, attention or uncertainty?
2024, International Review of Financial Analysis
Keyword-based measures purporting to reflect investor sentiment, attention or uncertainty have increasingly been used to model stock market behaviour. We investigate and shed light on the narrative reflected by Google search trends (GST) by constructing a neutral and general stock market-related GST index. To do so, we apply elastic net regression to select investor relevant search terms using a sample of 77 international stock markets. The index peaks around significant events that impacted global financial markets, moves closely with established measures of market uncertainty and it is predominantly correlated with uncertainty measures in differences, implying that GST reflect an uncertainty narrative. Returns and volatility for developed, emerging and frontier markets widely reflect changing Google search volumes and relationships conform to a priori expectations associated with uncertainty. Our index performs well relative to existing keyword-based uncertainty measures in its ability to approximate and predict systematic stock market drivers and factor dispersion underlying return volatility both in-sample and out-of-sample. Our study contributes to the understanding of the information reflected by GST, their relationship with stock markets and points towards generalisability, thus facilitating the development of further applications using internet search data.
The US banking crisis in 2023: Intraday attention and price variation of banks at risk
2023, Finance Research Letters
After interest rate hikes by the FED, the market value of long-duration assets has declined, which in March of 2023 led to a distress of banks subject to a sudden increase in deposit withdrawals. First, Silicon Valley Bank and Signature Bank were subject to such runs and were taken over by the FDIC, while First Republic Bank by JPMorgan Chase in May of 2023. We date the crisis-period and show that increased Twitter-based attention is related to an increased price variation of banks at risk. Our results imply that during crisis periods, pricing can be improved by incorporating attention-based measures.
Media attention and corporate greenwashing behavior: Evidence from China
2023, Finance Research Letters
This study examines the impact of media attention on corporate green drift behavior and the mediating role of executive risk preferences in it. This study examines the green innovation of firms using A-share listed companies from 2011 to 2021 as the research sample. It is empirically tested that media attention significantly inhibits corporate greenwashing behavior, and executive risk preferences positively reinforce this inhibitory effect. The findings provide a new perspective for preventing corporate greenwash initiatives, which, together with a moderate increase in executive risk-taking, has practical implications for promoting the high-quality development of corporate green ecological civilization.
EU Climate Change News Index: Forecasting EU ETS prices with online news
2023, Finance Research Letters
Carbon prices have been rapidly increasing in the EU since 2018 and accurate forecasting of EU Emissions Trading System (ETS) prices has become essential. This paper proposes a novel method to generate alternative predictors for daily ETS price returns using relevant online news information. We devise the EU Climate Change News Index by calculating the term frequency–inverse document frequency (TF–IDF) feature for climate change-related keywords. The index is capable of tracking the ongoing debate about climate change in the EU. Finally, we show that incorporating the index in a simple predictive model significantly improves forecasts of ETS price returns.
Introspecting predictability of market fear in Indian context during COVID-19 pandemic: An integrated approach of applied predictive modelling and explainable AI
2021, International Journal of Information Management Data Insights
Citation Excerpt :
Liu et al. (2021) argued fear sentiment owing to COVID-19 pandemic augmented the possibility of market crash. In general media coverage, sentiments, etc. have been observed to possess some explanatory capacity in governing the volatility of financial markets Guidolin & Pedio (2021). The present work resorts to Google Search Volume Index (GSVI) (Swamy, Dharani & Takeda, 2019; Afkhami, Ghoddusi & Rafizadeh, 2021) on different keywords to extract the cognate sentiment effectively.
Financial markets across the globe have seen rapid volatility and uncertainty owing to scary and disruptive impacts of COVID-19 pandemic. Mayhem wrecked by frequent lockdowns, curfews, emergencies, etc. has stoked the high quantum of chaotic movement in equity markets and resulted in perplexed investor behaviour. It, therefore, is of paramount practical relevance to measure predictability of market fear at such a crucial juncture of time. Market fear can effectively be measured in terms of implied and historic volatility of equity markets. The present study chooses India VIX and 20-day rolling standard deviation of NIFTY returns to account for implied and historic volatility respectively during the ongoing COVID-19 timeline. Pertinent macroeconomic constructs, technical indicators and Google search volume index on meaningful keywords have been chosen as raw explanatory features for inspecting predictability. Boruta feature selection methodology has been used in a supervised manner to select significant features. State-Of-The-Art machine and deep learning algorithms namely Gradient Boosting (GB), Extra Tree Regression (ERT), Deep Neural Network (DNN), Long Short Term Memory Network (LSTM) are then used on processed feature set to scrupulously evaluate the quantum of predictability of said assets. The integrated predictive frameworks have been subjected to a battery of numerical and statistical checks to draw inferences. Additionally, Explainable AI frameworks have been used to analyse the nature of influence of respective features. Findings indeed suggest that despite exhibiting high degree of volatile traits, both India VIX and historic volatility can be predicted utilizing the proposed architectures effectively and serve practical actionable insights.
The Financial Impact of Chatgpt for the Higher Education Industry in the U.S
2023, SSRN

View all citing articles on Scopus

: We thank three anonymous referee for constructive feedbacks and Alessandra Poli for excellent research assistance.

: JEL codes: C53, C58, G17.

View full text

Media Attention vs. Sentiment as Drivers of Conditional Volatility Predictions: An Application to Brexit

Highlights

Abstract

Introduction

Section snippets

Data

Methodology

In-sample forecasting results

Conclusion

Author's Statement (on Behalf of All Authors)

Finance Research Letters

Finance Research Letters

Journal of Financial Economics

European Economic Review

International Review of Financial Analysis

Economics Letters

Impact of Brexit vote on the London stock exchange: A sectorial analysis of its volatility and efficiency

Finance Research Letters

Investor attention and stock market volatility

Review of Financial Studies

Is all that talk just noise? The information content of Internet stock message boards

Journal of Finance

Policy uncertainty and international financial markets: the case of Brexit

Applied Economics