Chapter 4 Intangible Cost Estimation by Twitter Sentiment Event Study 46
two contributions. Firstly, with the analysis of public enterprises, we demonstrate our approach, and we confirm high correlations (Correlation Coefficient: +0.8) between stock price data and Tweet sentiment data in short-term (3 days before and after the Event Day) by analyzing both data. Secondly, we apply this method to the non-public organization not having stock price data, in order to prove the applicability of our proposed approach.
Chapter 4 Intangible Cost Estimation by Twitter Sentiment Event Study 47
negative opinions and evaluating existing image and impacts on the targeted en-tity. Recently, social media marketing and communication have been a critical do-main of marketing department. Secondly, Twitter allows to use only 140-character words, and it realizes “readily usable” and “real-time update” features. We think these features reflect the response against victims’ organization quickly.
4.2.3 Data Simlirality
Both data have pros and cons, but, we would like to show the similarity of both data.
Firstly, as we mentioned, a common characteristic of them is elaborating the popularity and corporate value of public companies.
Secondly, both data has a close relationship. Paper [157] reported that they could predict the stock price by using Twitter sentiment analysis and the success ratio was 87.6% because of a strong correlation between both data. According to the article [158], European hedge fund developed algorithm trading system based on this research, and they achieved remarkable investment performance. In addition to this, the market also considers that negative impact on Twitter is one of the very critical factors to estimate stock price because recent flash trade (HFT:
High-Frequency Trade) algorithm referred the contents of Twitter. For example, when the Twitter account of Associated Press was hacked, and hacker released fake news on this compromised Twitter, it had substantial negative impact on NASDAQ or Dow Jones Industrial Average [159].
Since Twitter sentiment has been a significant factor of stock price, several Japanese companies started to apply the sentiment value for the stock price esti-mation. For example, NTT Data launched the service called “Twitter Sentiment Index” by using Twitter sentiment data for financial market in 2014 [160]. In addi-tion to this, NRI had an empirical study of natural language analysis to investment judgment in 2017 [161].
Chapter 4 Intangible Cost Estimation by Twitter Sentiment Event Study 48
4.2.4 Data Difference
In this section, we will compare both data from five perspectives.
Comparison 1 : Applicability
Firstly, Twitter data has more applicable to more organization. Since only public companies have the stock price data, and the traditional event study method can evaluate only public companies. However, in the proposed method, all entities, including not only public companies but also organizations not having stock data such as private companies, government agencies, and non-profit organizations, will be evaluated, as long as Twitter data are available for them. Therefore, from the applicability perspective, the Twitter dataset is more applicable to many situations.
Comparison 2 : Users Amount
From user amount perspective, we believe that larger population reflects more diversified opinions and the population of Twitter users is greater than one of stock traders. According to statistical data in Japan, although there are 14 million individual stock traders [162, 163], Twitter has 40 million active users [164]. In addition to this, the people who reflect opinions via stock are only stockholders, and actual influential stakeholders are limited. It means that Tweet data can reflect broader views for the security incident.
Comparison 3 : Incentive of Stakeholders
We think Twitter data has more honest opinions against information breach be-cause the incentive of stakeholders is different.
The primary purpose of stock trading is not an evaluation of corporate value or a reflection of opinions, but gaining margin or capital gains by trading. It means that the stock price does not always reflect the actual opinion of the security incident. For example, some strategic traders may purchase stocks after security
Chapter 4 Intangible Cost Estimation by Twitter Sentiment Event Study 49
breach disclosure because they get them with a small price to sell them at a higher price in the future. In another case, some traders may keep stocks because security breach is temporary events and it is not so impressive from long-term perspectives.
Therefore, the stock price may not reflect the impact of valuation.
On the contrary, the primary purpose of Twitter is communication, and Twitter users use these services to publish, spread and gather the news, valuable informa-tion, and individual opinion. In another word, active Twitter users do not tweet their opinions by monetary incentives, and all tweets related to incident response is honest opinions against the organization and incidents.
Comparison 4 : Side Effect Elimination
We consider that our proposed methods can eliminate unrelated data or another event effects from Twitter data if several incidents or events related to reputation risks are handled simultaneously, and the analysts can focus on the analysis of the particular event deeply. In a multi-incident situation, it is difficult to distinguish one incident impact from the others in stock price data. However, in Twitter dataset, analysts can pick up relevant data with keyword search and filter out irrelevant data from the dataset. This side effect elimination is one of the unique capabilities our proposed methods, and Twitter dataset has.
Comparison 5 : Real-Time Evaluation
As a final point, we consider that Twitter data has real-time evaluation capability rather than stock data. Stock data tends to be usually delayed to reflect the market opinions to actual price because stock price will be decided by matching of sell order and buy order. In addition to this, real-time evaluation by stock data is only available when stock exchange markets open. In another word, the stock price is not useful to evaluate the reputation on the weekend or after closing markets, although negative reputation spread in anytime. On the contrary, Twitter data can reflect opinions in real-time because publishing tweets requires no prerequisite
Chapter 4 Intangible Cost Estimation by Twitter Sentiment Event Study 50
process and the analysts can use these data as long as Twitter users comment on them actively.
Also, Twitter data is also effective from risk communication and social media communications domain, because modern corporate communication utilizes social media and many important announcements also tends to be published on Twit-ter. By using the proposed method, we consider that PR (Public Relations) or Marketing department in each victimized organizations can utilize this approach to analyze the impact of each announcement.
4.2.5 Dataset Sumamry
In this section, we discussed the similarity and difference of both data. In Section 4.6, based on the similarity assumption, we have several case studies by applying the proposed approach. We would like to notice that we will only verify the similarity of both data, and also the effectiveness of “Applicability”. We note that the impact analysis caused by the other data differences will be discussed as a future work.