Earticle

초록

영어

Since its emergence in 2008, Bitcoin has been experiencing continuous growth in popularity worldwide. According to the Google Trends, Bitcoin search results burgeoned more than 10 folds just within the last 12 months. Nowadays, Bitcoin is becoming more and more acceptable in online transactions. Most of cryptocurrencies, including Bitcoin, are traded on the web. Due to its innately cybernetic characteristics, Bitcoin has allowed user opinion on the Internet communities to gain the power to act as a miniature mirror that reflects user response to the ever-growing cryptocurrency market. “Virtual Currency, Tangible Return: Portfolio Diversification with Bitcoin (M. Brière, 2015)” concluded that high volatility (-41.78 to 136.72%) of cryptocurrencies matched with the characteristics of high-risk high-return investments. In addition, notable differences between stock and cryptocurrency markets encourage volatile return on cryptocurrency investments. Firstly, the trading hour of the cryptocurrency market is much longer than that of the stock market, which makes it a restless mechanism. Secondly, the stock market is geographically segregated, while the cryptocurrency market is internationally accepted. Thirdly, the outstanding numbers of stocks and cryptocurrencies are controlled differently. Simply said, the outstanding number of stocks is decided by the company through issuance, whereas the outstanding number of cryptocurrencies is heavily reliant on the mining speed of a specific cryptocurrency. This very characteristic affects the power of the issuer. Fourthly, the most distinctive difference between stocks and cryptocurrencies is that stock prices are dependent on tangible assets while cryptocurrency prices are dependent on perceived market value. Consequently, cryptocurrency prices are more susceptible to investors’ perceptions compared to stock prices. Because only two factors, 1) perceived market value and 2) principles of supply & demand, heavily affect cryptocurrency prices, the following paper seeks to examine the effects of online investors’ sentiment on the cryptocurrency market. As stated by Robert Plutchik, the eight types of emotions – joy, acceptance, anticipation, surprise, fear, sorrow, disgust, and anger – are categorized either into a positive or negative domain, with an exception of surprise. Joy, acceptance and anticipation are positive emotions, while fear, sorrow, disgust, and anger are negative emotions; surprise falls into the category of neutral emotion. This paper looks towards identifying which emotion within each category poses the most influence on Bitcoin market price, by cross-comparing Bitcoin price history and the percentage of respective sentimental user comments and replies on the online forum. Among the eight types of emotions proposed by Robert Plutchik, there are two specific emotions that stand out from all others, which are ‘anticipation’ and ‘fear’. While other emotions show a human response to the past event, the two above-stated emotions convey the sense of uncertainty towards the upcoming event. There are a few previous literature on cryptocurrency sentiment analysis. Firstly, in “Predicting Bitcoin Price Fluctuations with Twitter Sentiment Analysis (E. Stenqvist, 2017),” the researchers analyze Bitcoin-related tweets for sentiment fluctuations that could indicate price change. The authors implement the sentiment classification of 1) positive, 2) neutral, and 3) negative, which is different from the research. As a result, the paper showed a sentiment change threshold of 2.2% with a 79% accuracy. This study provided two important points for this research: 1) there is a relationship between sentiment and cryptocurrency price fluctuations, 2) the lag effect exists. Secondly, the papers, “Algorithmic Trading of Cryptocurrency Based on Twitter Sentiment Analysis (S. Colianni, 2015)” and “Cryptocurrency Price Prediction Using News and Social Media Sentiment (C. Lamon, 2017),” sought to prove the relationship between public sentiment and cryptocurrency price fluctuations. Similar to the first literature, they both used the sentiment classification of 1) positive, and 2) negative. Moreover, the finding concurs with the previous one that there is a relationship between sentiment and cryptocurrency fluctuations. The main takeaway from the literature review on current cryptocurrency sentiment analysis is that they only focus on positive and negative nature of the comments. This research is unique in a sense that it emphasizes the sentiments’ conveyance of uncertainty. Furthermore, this paper seek to stand out from previous literature on stock market sentiment analysis. As shown in “Twitter Mood Predicts the Stock Market (J. Bollen, 2011),” stock market sentiment analysis researches aimed to show the relationship between public sentiment and stock price fluctuations. Here, the sentiment classification applied were 1) calm, 2) alert, 3) sure, 4) vital, 5) kind, and 6) happy. However, there were not any firm reasoning behind choosing such lexicons to distinguish sentiment categories. The sentimental class of calm yielded 86.7% direction accuracy of the fluctuation with baseline prediction of 73.3%. The result, in turn, indicates sentiment association in the stock market. This strengthens the outcome of this research. As for an expected outcome, a number of hypotheses were deduced based on preliminary research. According to the theoretical background, hypothesizing that the emotions of anticipation and fear will hold a higher ground over other emotions when affecting the Bitcoin market is reasonable. Consequently, the following hypotheses were developed: H1a: Anticipation is positively associated with Bitcoin return. H1b: Fear is negatively associated with Bitcoin return. H2a: Anticipation is positively associated with Bitcoin price volatility. H2b: Fear is positively associated with Bitcoin price volatility. To accomplish the proposed purpose, the paper crawled all user comments and replies from the online Bitcoin forum. Through the data analysis process, the paper analyzed the data and categorized it according to the extent of relevance to the two respective emotions conveying uncertainty (anticipation and fear) proposed by Robert Plutchik. Subsequently, the paper cross-compared the relationship between Bitcoin price history and the quantity of respective sentimental user comments and replies in order to test the level of influence of anticipation and fear on the Bitcoin price fluctuation. In order to control the difference between the comment generating rates between day and night, percentage, rather than mere number, of comments are used to identify the degree of sentimental concentration. Hence, the two independent variables were: 1) percentage of comments and replies with anticipation, and 2) percentage of comments and replies with fear. The two dependent variables for this paper were: 1) Bitcoin return calculated in comparison with the previous collection interval, and 2) Bitcoin price volatility calculated in comparison with the previous collection interval. A number of controlled variables were implemented to maximize the influence of public sentiment. Concentration on specific cryptocurrency and promotional events that may attract more investors were carefully addressed. Also, spikes in Google Trends is identical to the trading volume spikes which result in sudden fluctuations; hence, Google Trend spikes were exempt for data collection. The following variables were considered: 1) total number of comments in 6-hour interval, 2) ICO of a new cryptocurrency, 3) delisting of a cryptocurrency, 4) promotional events conducted at Binance Exchange. Throughout the data collection process, extensive data was collected from the most popular cryptocurrency community. Sentimental user comments and replies accumulated in the largest Bitcoin community, Bitcoin Talk, was crawled for sentimental analysis. Similarly, most popular US Bitcoin exchange, Binance, was used to collect past Bitcoin market price history. Both sentimental user reactions and Bitcoin price history database composed of a minimum of three-month period. For the data analysis process, anticipation and fear in user response was applied in identifying the correlation to the Bitcoin market. Then, a consistent number of keywords that represent each of the two emotions was chosen after extensive preliminary data mining. Following this, the percentage of each emotion was calculated in order to deduce the relationships between 1) anticipation and Bitcoin return, and 2) fear and Bitcoin return. In addition, the relationships between anticipation/fear and Bitcoin price volatility was examined to further analyze investors’ sentiment on cryptocurrency market. By evaluating the degree of correlation and the percentage of each emotion within the fluctuation, the paper identifies whether emotional user comments and replies have any association with the Bitcoin return/volatility. For sentiment classification, two sentimental groups, anticipation and fear, was used. In reference to the SentiWordNet 3.0, words expressing uncertainty was selected (S. Baccianella 2010). To complete data analysis process, time-series regression model was implemented. Furthermore, after defining the time-sensitive relationship between public sentiments and Bitcoin indices, Granger Causality was used to identify the cause and effect relationship between the variables. The ultimate goal of this research is to develop and test a prediction model for Bitcoin prices. In order to do so, evident cause-and-effect relationship has to be established between anticipation/fear-related public sentiments and Bitcoin return. After establishing the valid relationship, the concept of machine learning, especially deep neural network (DNN) will be explored to structure the prediction model based on public sentiments.

목차

Abstract

키워드

저자

• Kyuho Han [ Korea Advanced Institute of Science and Technology (KAIST) ]

참고문헌

발행기관

간행물

표지보기 관심저널 등록

• 간행물명

  한국경영정보학회 정기 학술대회 [KMIS Conference]

• 간기

  반년간

• 수록기간

  1990~2025

• 십진분류

  KDC 325 DDC 658