Tweets and Trades: the Information Content of Stock Microblogs

• DOI: http://doi.org/10.1111/j.1468-036X.2013.12007.x
• Published date: 01 November 2014
• Date: 01 November 2014

Tweets and Trades: the Information

Content of Stock Microblogs

Timm O. Sprenger, Andranik Tumasjan,

Philipp G. Sandner and Isabell M. Welpe

Technische Universität München, TUM School of Management, Germany

E-mail: timm.sprenger@gmail.com

Abstract

Microblogging forums (e.g., Twitter) have become a vibrant online platform for

exchanging stock‐related information. Using methods from computational

linguistics, we analyse roughly 250,000 stock‐related messages (so‐called tweets)

on a daily basis. We find an association between tweet sentiment and stock returns,

message volume and trading volume, as well as disagreement and volatility. In

contrast to previous related research, we also analyse the mechanism leading to an

efficient aggregation of information in microblogging forums. Our results

demonstrate that users providing above average investment advice are retweeted

(i.e., quoted) more often and have more followers, which amplifies their share of

voice.

Keywords: Twitter,microblogging,stock market,investor sentiment,text classifica-

tion,computational linguistics

JEL classification: G12, G14

1. Introduction

Scholars and practitioners alike increasingly call attention to the popularity of online

investment forums among investors and other financial professionals (Antweiler and

Frank, 2004; BusinessWeek, 2009). Stock microblogging, mostly based on the social

networking service Twitter (www.twitter.com), has recently been at the forefront of this

development. Some commentators have even described the conversations on this

platform as ‘the modern version of traders shouting in the pits’(BusinessWeek, 2009).

Twitter is a microblogging service allowing users to publish short messages with up to

140 characters, so‐called ‘tweets’. These tweets are visible on a public message board

of the website or through various third‐party applications. Users can subscribe to

We thank John Doukas (the editor of this Journal) and an anonymous referee for insightful

comments and suggestions that helped to improve this article. Correspondence: Timm O.

Sprenger.

European Financial Management, Vol. 20, No. 5, 2014, 926–957

doi: 10.1111/j.1468-036X.2013.12007.x

© 2013 John Wiley & Sons Ltd

(i.e., ‘follow’) a selection of favorite authors or search for messages containing a specific

key word (e.g., a stock symbol). The public timeline has turned into an extensive real‐time

information stream of currently more than 340 million messages per day generated by

roughly 140 million active users (ZDNet, 2012). Many of these messages are dedicated to

the discussion of public companies and trading ideas. As a result, there are investors who

attribute their trading success to the information they find on social media websites

and Twitter‐based trading systems have been developed by financial professionals to

alert users of sentiment‐based investment opportunities (Jordan, 2010), and by academic

researchers to predict break‐points in financial time series (Vincent and Armstrong,

2010).

As yet, there is only little research examining whether and how microblogging

messages are related to financial indicators. For instance, Sprenger et al. (2012)

investigate how different company‐specific news events published on Twitter (e.g.,

corporate governance or legal issues) are related to S&P 500 stock prices. Their

comprehensive study shows that the content of Twitter messages provides valuable

information with regard to the effects of different stock‐related news types on company

financial indicators. In a related vein, Sprenger and Welpe (2011) demonstrate that joint

company mentions in Twitter messages predict their stocks’comovement. Moreover,

their study indicates that this measure of company relatedness can also be used to

delineate homogenous industry groups.

A few recent studies have made a first step in exploring whether the content of Twitter

may help predict macroeconomic market indicators. For instance, Bollen et al. (2011)

investigated whether collective mood states on Twitter are related to the value of the Dow

Jones Industrial Average (DJIA). They find that certain mood states are indeed predictive

of the DJIA closing values. In a similar vein, Zhang et al. (2010) explored the relationship

between hope and fear on the one hand and the Dow Jones, NASDAQ, and S&P 500 on

the other hand. Their results indicate that the level of tweet emotionality was significantly

related to all three aggregated indicators. However, while these studies offer a first

indication of the relationship between tweets and aggregated financial indicators, they are

also limited in a number of ways. First, both studies use randomised subsamples of all

available tweets of the Twitter message stream. Since the majority of these messages may

arguably not be related to stock market topics, we cannot infer whether the stock‐specific

information contained in tweets is indeed associated with these indicators. Second, both

studies only explored the relationship between aggregate sentiment measures and

aggregate stock market indices, which does not allow us to draw conclusions on whether

the information contained in stock‐related messages is related to the performance of

individual stocks. Since Das and Chen (2007) found the relationship between aggregated

sentiment and index returns to be much stronger than the correlation for individual stocks,

a more conservative approach focusing on the specific domain of stock‐related messages

and their relationship with the market prices of publicly traded companies (rather than

relying on aggregated indices) is needed. Third, and most importantly, there is as yet no

research investigating the mechanism underlying the link between social media message

sentiment (e.g., tweet sentiment) and market prices. As a result, we do not know how

information diffuses in social media in general, and Twitter microblogging in particular,

leading to efficient information processing.

Our study addresses these limitations of prior related work and makes the following

three main contributions to the literature. First, unlike previous related research (e.g.,

Bollen et al., 2010; Zhang et al., 2010), we analyse only microblogging messages with a

© 2013 John Wiley & Sons Ltd

Tweets and Trades 927

direct reference to the stock market (i.e., we analyse only explicit stock microblogging

messages rather than all available Twitter messages as in prior studies) which allows us to

determine the predictive validity of stock microblogs without ‘noise’that is unrelated to

the stock market, as has been the case in prior studies (e.g., Bollen et al., 2010; Zhang

et al., 2010).

Second, to the best of our knowledge, our study is the first to comprehensively explore

the information content of stock microblogs with respect to individual stocks rather than

aggregate stock market indices. In contrast to related previous studies, this study is able to

go beyond the analysis of relatively simple measures of online activity (e.g., message

volume or word counts), instead leveraging an innovative methodology from

computational linguistics to evaluate the actual message content and sentiment.

Moreover, this study replicates and extends similar research in the context of internet

message boards without some of the previous methodological limitations (e.g., sample

selection, timeframe; Antweiler and Frank, 2004). We analyse a more comprehensive set

of stocks over the course of 6 months with fairly stable financial market activity.

Third, in contrast to the existing research that has merely shown a correlation of online

message content with financial market indicators (e.g., Antweiler and Frank, 2004; Bollen

et al., 2010; Zhang et al., 2010), our study goes beyond this correlational evidence by

providing an explanation for the underlying mechanism that leads to the efficient

aggregation of information in stock microblogging forums. Unlike these previous studies,

we explicitly exploit the structure of the microblogging forum Twitter,which allows us to

empirically explore theories of social influence concerning the diffusion and processing of

information in the context of a financial community, which is not possible in traditional

message boards (e.g., Antweiler and Frank, 2004).

In particular, the present study takes the following steps in extending the existing

literature on the relationship between social media content and stock performance using

the microblog forum Twitter. First, for comparability with prior related research (e.g.,

Antweiler and Frank, 2004), our study examines the relationship between the most

important and heavily studied market features return, trading volume, and volatility and

the corresponding tweet features message sentiment (i.e., bullishness),

1

message volume,

and the level of agreement among postings. Second, we empirically explore the

mechanism behind the efficient aggregation of information in microblogging forums.

Specifically, we investigate the association between the quality of investment advice and

the level of mentions, the rate of retweets, and the authors’followership.

2. Related Work and Research Questions

2.1 Introduction to the research of online stock forums

One of the most intriguing sources of unofficial and qualitative information is the vast

amount of user‐generated content online. In the context of the stock market, internet

forums dedicated to financial topics, such as internet stock message boards like Yahoo!

Finance, deserve special attention. A number of previous studies have investigated the

relationship between stock message boards and financial markets. Wysocki (1998) was

the first to investigate internet stock message boards. For the 50 most frequently discussed

1

We use the terms sentiment and bullishness interchangeably.

© 2013 John Wiley & Sons Ltd

928 T.O. Sprenger, A. Tumasjan, P. G. Sandner and I. M. Welpe