Twitter Mining for Discovery,Prediction and Causality: Applications and Methodologies

Twitter has found substantial use in a number of settings. For example, Twitter played a major role in the ‘Arab Spring’ and has been adopted by a large number of the Fortune 100. All of these and other events have led to a large database of Twitter tweets that has attracted the attention of a number of companies and researchers through what has become known as ‘Twitter mining’ (also known as ‘TwitterMining’). This paper analyses some of the approaches used to gather information and knowledge from Twitter for Twitter mining. In addition, this paper reviews a number of the applications that employ Twitter Mining, investigating Twitter information for prediction, discovery and as an informational basis of causation. Copyright © 2015 John Wiley & Sons, Ltd.     

Explaining stock markets' performance during the COVID-19 crisis: Could Google searches be a significant behavioral indicator?

The purpose of this study is to examine the impact of the pandemic on the performance of stock markets, focusing on the behavioral influence of the fear due to COVID-19. Using a data set of 10 developed countries during the period December 31, 2019, to September 30, 2020, we examine the impact of COVID-19 on the performance of the stock markets. We incorporate the impact of the COVID-19 pandemic using the following variables: (a) the number of new COVID-19 cases, which was widely used as the main explanatory variable for market performance in early financial studies, and (b) a Google Search index, which collects the number of Google searches related to COVID-19 and incorporates the health risk and the fear of COVID-19 (the higher the number of searches for Covid terms, the higher the index value, and the higher the fear index). We employ our input into an EGARCH(1,1,1) model, and the findings show that the Google Search index enables us to draw statistically significant information regarding the impact of the COVID-19 fear on the performance of the stock markets. On the other hand, the variable of the number of new COVID-19 cases does not have any statistically significant influence on the performance of the stock markets. Google searches could be a useful tool for supporters of behavioral finance, scholars, and practitioners.     

An encyclopedia for stock markets? Wikipedia searches and stock returns

We present empirical evidence that collective investor behavior can be inferred from large-scale Wikipedia search data for individual-level stocks. Drawing upon Shannon transfer entropy, a model-free measure that considers any kind of statistical dependence between two time series, we quantify the statistical information flow between daily company-specific Wikipedia searches and stock returns for a sample of 447 stocks from 2008 to 2017. The resulting stock-wise measures on information transmission are then used as a signal within a hypothetical trading strategy. The results evidence the predictive power of Wikipedia searches and are in line with the previously documented notion of buying pressure revealed by online investor attention and the trading patterns of retail investors.     

How do the content,format, and tone of Twitter-based corporate disclosure vary depending on earnings performance?

Using 86,891 tweets, from the official corporate Twitter accounts of 715 unique firms, this study examines whether and how managers strategically attract and distract investors’ attention from corporate news through Twitter. We find that firms with good earnings news use Twitter to post more earnings-related information directly, whereas firms with bad earnings news post more non-earnings-related information on Twitter. We further find that depending on earnings performance firms strategically choose the format of tweets (qualitative or quantitative) and the tone of earnings tweets (positive or negative) to attract investors’ attention to good news or distract investors’ attention from bad news. Our results are robust to difference-in-differences (DID), alternative sample periods, and different variable specifications. Our findings provide empirical evidence for investors and regulators regarding current practices in corporate information on Twitter.     

Investor Information Demand: Evidence from Google Searches Around Earnings Announcements

The objective of this study is to investigate factors that influence investor information demand around earnings announcements and to provide insights into how variation in information demand impacts the capital market response to earnings. The Internet is one channel through which public information is disseminated to investors and we propose that one way that investors express their demand for public information is via Google searches. We find that abnormal Google search increases about two weeks prior to the earnings announcement, spikes markedly at the announcement, and continues at high levels for a period after the announcement. This finding suggests that information diffusion is not instantaneous with the release of the earnings information, but rather is spread over a period surrounding the announcement. We also find that information demand is positively associated with media attention and news, and is negatively associated with investor distraction. When investors search for more information in the days just prior to the announcement, preannouncement price and volume changes reflect more of the upcoming earnings news and there is less of a price and volume response when the news is announced. This result suggests that, when investors demand more information about a firm, the information content of the earnings announcement is partially preempted.     

Investor attention and global market returns during the COVID-19 crisis

The financial market response to the COVID-19 pandemic provides the first example of a market crash instigated by a health crisis. As such, the crisis provides a unique setting in which to examine the market response to changes in investor attention. We utilise Google search volume (GSV) as a proxy for investor attention. GSV for the “coronavirus” keyword increases markedly from late-February and peaks in mid-March before declining substantially. Our results are broadly consistent with Da, Engelberg, and Gao (2015), indicating that GSV is primarily a proxy for the attention of retail investors and confirming that investor attention negatively influences global stock returns during this crisis period. A rise in the number of internet searches during the COVID-19 crisis induces a faster rate of information flow into financial markets and so is also associated with higher volatility. The identified relationships are economically and statistically significant even after controlling for the number of COVID-19 cases and macroeconomic effects. Increases in GSV have less impact on government bond yields where the limited role of GSV is likely due to lower participation of retail investors. The results suggest that, rather than searching for information on potential stocks to buy (Barber & Odean, 2008), retail investors are searching for information to resolve uncertainty about household FEARS (Da et al., 2015) during the COVID-19 crisis.     

Internet search volumes of UK banks during the crisis: The role of banking structure and business model

This paper investigates the sensitivity of Google search volumes to the global financial crisis, using a large sample of UK banks. We find that abnormal volumes of searches on bank names are a timely indicator of credit risk, as measured by credit default swap rates, and are strongly related to crisis indicators. Search volumes for retail, private, and universal banks react more strongly than search volumes for other banks, suggesting that depositor concerns drive search volumes. Search volumes for internet banks and for European banks operating in the UK through branches are especially sensitive to crises.     

Twitter activity,investor attention,and the diffusion of information

We examine the impact of Twitter attention on stock prices by examining over 21 million company‐specific tweets over a 5‐year period. Through a quasi‐natural experiment identifying official Twitter outages, we find that Twitter influences stock trading, especially among small, less visible securities primarily traded by retail investors. In addition, we determine that Twitter activity is associated with positive abnormal returns and when tweets occur in conjunction with traditional news events, more information is spread to investors. Finally, we show that retail investor activity drives the Twitter effect as institutional investors less actively trade the affected stocks.     

Information demand during the COVID-19 pandemic

We investigate the demand for financial information during the initial months of the COVID-19 pandemic. Using Google search data for individual stocks, we show that the Abnormal Google Search Volume Index declined significantly between March and June of 2020. We find a similar effect around earnings announcements dates, which confirms that the demand for financial information by retail investors declined during the pandemic. Our results are indicative of potentially important consequences for information diffusion, price discovery and market efficiency under extreme uncertainty. We discuss possible explanations for these results.     

DO INVESTORS CARE ABOUT PRESIDENTIAL COMPANY‐SPECIFIC TWEETS?

When the president of the United States tweets, do investors respond? We analyze the impact of tweets from President Donald J. Trump's official Twitter accounts from November 9, 2016 to December 31, 2017 that include names of publicly traded companies. We find that these tweets move company stock prices and increase trading volume, volatility, and institutional investor attention, with a stronger impact before the presidential inauguration. There is some evidence that the initial impact of the presidential tweets on stock prices is reversed in the next few trading days.     

Tweets and Trades: the Information Content of Stock Microblogs

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.     

Wisdom of Crowds: Cross‐Sectional Variation in the Informativeness of Third‐Party‐Generated Product Information on Twitter

This paper examines whether third‐party‐generated product information on Twitter, once aggregated at the firm level, is predictive of firm‐level sales, and if so, what factors determine the cross‐sectional variation in the predictive power. First, the predictive power of Twitter comments increases with the extent to which they fairly represent the broad customer response to products and brands. The predictive power is greater for firms whose major customers are consumers rather than businesses. Second, the word‐of‐mouth effect of Twitter comments is greater when advertising is limited. Third, a detailed analysis of the identity of the tweet handles provides the additional insights that the predictive power of the volume of Twitter comments is dominated by “the wisdom of crowds,” whereas the predictive power of the valence of Twitter comments is largely attributable to expert comments. Furthermore, Twitter comments not only reflect upcoming sales, but also capture an unexpected component of sales growth.     

Google Internet search activity and volatility prediction in the market for foreign currency

I study whether evolution in the number of Google Internet searches for particular keywords can predict volatility in the market for foreign currency. I find that data on Google searches for the keywords economic crisis + financial crisis and recession has incremental predictive power beyond the GARCH(1,1). These results support the mixture of distributions hypothesis in that volatility is linked to the stochastic rate at which information flows into the marketplace. These results also demonstrate the potential for Google to become a storehouse of information for financial markets.     

Google searches and stock returns

We investigate whether data from Google Trends can be used to forecast stock returns. Previous studies have found that high Google search volumes predict high returns for the first one to two weeks, with subsequent price reversal. By using a more recent dataset that covers the period from 2008 to 2013 we find that high Google search volumes lead to negative returns. We also examine a trading strategy based on selling stocks with high Google search volumes and buying stocks with infrequent Google searches. This strategy is profitable when the transaction cost is not taken into account but is not profitable if we take into account transaction costs.     

Retail investor attention and the limit order book: Intraday analysis of attention-based trading

We are the first to examine how intraday changes in retail investor attention, measured by hourly Google searches, affect trading activity and informativeness of trades. High levels of Google search activity are followed in the next hour by more intensive trading in all stocks. The increased trading activity is initiated by retail investors as evidenced by the reduced size of new orders. After googling a company, retail investors do not become informed in the traditional sense; rather, they act as noise traders, who mistake noise for information, as their orders are picked off by truly informed traders.     

The Role of Social Media in the Capital Market: Evidence from Consumer Product Recalls

We examine how corporate social media affects the capital market consequences of firms’ disclosure in the context of consumer product recalls. Product recalls constitute a “product crisis” exposing the firm to reputational damage, loss of future sales, and legal liability. During such a crisis it is crucial for the firm to quickly and directly communicate its intended message to a wide network of stakeholders, which, in turn, renders corporate social media a potentially useful channel of disclosure. While we document that corporate social media, on average, attenuates the negative price reaction to recall announcements, the attenuation benefits of corporate social media vary with the level of control the firm has over its social media content. In particular, with the arrival of Facebook and Twitter, firms relinquished complete control over their social media content, and the attenuation benefits of corporate social media, while still significant, lessened. Detailed Twitter analysis confirms that the moderating effect of social media varies with the level of firm involvement and with the amount of control exerted by other users: the negative price reaction to a recall is attenuated by the frequency of tweets by the firm, while exacerbated by the frequency of tweets by other users.     

Wikipedia pageviews as investors’ attention indicator for Nasdaq

The attempt to measure investors’ mood to find an early indicator of financial markets has evolved and developed with the advancement of technology over the years. The first attempts were based on surveys, a long and expensive process. Nowadays, big data has made it possible to measure the investor’s mood accurately and almost entirely online. This paper analyzes the explanatory and predictive capacity of Wikipedia pageviews for the Nasdaq index. For this purpose, two econometric models have been developed. In both models, the explanatory variable is the number of Wikipedia visits, and the endogenous variable is Nasdaq index return. As an alternative to this approach, an algorithmic trading system has been developed. It uses Wikipedia visits as investment signals for long and short positions to check the predictability power of this indicator. It is determined that the volume of queries about Nasdaq companies is a statistically significant variable for expressing the evolution of this index. However, it has no predictive capacity. Keeping in mind the capacity of Wikipedia to exemplify Nasdaq trends, further studies should be conducted to determine how to make this indicator profitable.     

The social amplification of risk on Twitter: the case of ash dieback disease in the United Kingdom

It has long been recognised that the traditional media play a key role in representing risk and are a significant source of information which can shape how people perceive and respond to hazard events. Early work utilising the social amplification of risk framework (SARF) sought to understand the discrepancy between expert and lay perceptions of risk and patterns of risk intensification and attenuation with reference to the media. However, the advent of Web 2.0 challenges traditional models of communication. To date there has been limited consideration of social media within the SARF and its role in mediating processes of risk perception and communication. Against this backdrop, we focus on the social media platform Twitter to consider the social amplification of risk in relation to ash dieback disease (Hymenoscyphus fraxineus); a tree health issue that attracted intense media attention when it was first identified in the UK in 2012. We present an empirical analysis of 25,600 tweets in order to explore what people were saying about ash dieback on Twitter, who was talking about it and how they talked about it. Our discussion outlines the themes around which talk about ash dieback was orientated, the significance of users’ environmental ‘affiliations’ and the role of including links (URLs) to traditional media coverage. We utilise the notion of ‘piggybacking’ to demonstrate how information is customised in line with group/individual identities and interests and introduce the concept of the ‘frame fragment’ to illustrate how information is selected and moved around Twitter emphasising certain features of the messages. The paper affords a detailed consideration of the way in which people and organisations simultaneously appropriate, construct and pass on risk-relevant information. A conclusion is that social media has the potential to transform the media landscape within which the SARF was originally conceived, presenting renewed challenges for risk communication.     

Which popular predictor is more useful to forecast international stock markets during the coronavirus pandemic: VIX vs EPU?

This study mainly investigates which predictors (VIX or EPU index) are useful to forecast future volatility for 19 equity indices based on HAR framework during coronavirus pandemic. Out-of-sample analysis shows that the HAR-RV-VIX model exhibits superior forecasting performance for 12 stock markets, while EPU index just can improve forecast accuracy for 5 equity indices, implying that VIX index is more useful for most stock markets' future volatility during coronavirus crisis. The results are robust in recursive window method, alternative realized measures and sub-sample analysis; moreover, VIX index still contains the strongest predictive ability by considering kitchen sink model and mean combination forecast. Furthermore, we further discuss the predictive effect of VIX and EPU index before the coronavirus crisis. Our article provides policy makers, researchers and investors with new insights into exploiting the predictive ability of VIX and EPU index for international stock markets during coronavirus pandemic.     

Constructing a positive sentiment index for COVID-19: Evidence from G20 stock markets

The present study investigates the degree of market responses through the scope of investors' sentiment during the COVID-19 pandemic across G20 markets by constructing a novel positive search volume index for COVID-19 (COVID19⁺). Our key findings, obtained using a Panel-GARCH model, indicate that an increased COVID19⁺ index suggests that investors decrease their COVID-19 related crisis sentiment by escalating their Google searches for positively associated COVID-19 related keywords. Specifically, we explore the predictive power of the newly constructed index on stock returns and volatility. According to our findings, investor sentiment positively (negatively) predicts the stock return (volatility) during the COVID-19. This is the first study assessing global sentiment by proposing a novel proxy and its impacts on the G20 equity market.