When Bitcoin encounters information in an online forum: Using text mining to analyse user opinions and predict value fluctuation

Kim, Young Bin; Lee, Jurim; Park, Nuri; Choo, Jaegul; Kim, Jong Hyun; Kim, Chang Hun

doi:10.1371/journal.pone.0177630

Cited by 69 publications

(53 citation statements)

References 39 publications

(37 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Stenqvist et al [6] try to predict Bitcoin price (up/down) using sentiment analysis on Twitter, and report that the sentiment change over a 30-minute period is useful for predicting price movement of 2 hours later, resulting in an accuracy of 79%. Instead of performing sentiment analysis on all social media content posted, Kim et al [7] extract the hottest topics on a Bitcoin-related forum and define a time series score to represent the "strength" of each topic. While these scores are not significant in Granger causality tests, a deep learning model with these scores as inputs leads to prediction (for price and transaction volume) accuracies ranging from 50%+ to 80%+.…”

Section: Literature Reviewmentioning

confidence: 99%

Cryptocurrency Investing Examined

Liew¹,

Li²,

Budavàri³

et al. 2019

The JBBA

View full text Add to dashboard Cite

In this work we examine the largest 100 cryptocurrency returns ranging from 2015 to early 2018. We concentrate our analysis on daily returns and find several interesting stylized facts. First, principal components analysis reveals a complex daily return generating process. As we examine data in the most recent year, we find that surprisingly more than one principal component appears to explain the cross-sectional variation. Second, similar to hedge fund returns, cryptocurrency returns suffer from the "beta-in-the-tails" hidden risk. Third, we find that predicting cryptocurrency movements with machine learning and artificial intelligence algorithms is marginally attractive with variation in predictability power per crypto-currency. Fourth, lower volatile cryptocurrencies are slightly more predictable than more volatile ones. Fifth, evidence exists that efficacy of distinct information sets varies across machine learning algorithms, showing that predictability may be much more complex given a set of machine learning algorithms. Finally, short-term predictability is very tenuous, which suggests that near-term cryptocurrency markets are semi-strong form efficient and therefore, day trading cryptocurrencies may be very challenging.

show abstract

Section: Literature Reviewmentioning

confidence: 99%

Cryptocurrency Investing Examined

Liew¹,

Li²,

Budavàri³

et al. 2019

The JBBA

View full text Add to dashboard Cite

show abstract

“…The same data source (bitcointalk.org) has also been used in other topic modelling work [11]. Granger causality was applied to discovered topics to investigate whether there were relationships present between the occurrence of particular topics and statistics relating to Bitcoin.…”

Section: B Topic Modellingmentioning

confidence: 99%

“…Table II shows notable topics selected for their coherent cryptocurrency-related content. These topics have been manually labelled, as is common in topic modelling [11,18]. The most probable words in each topic are retrieved from the final point in the dataset and displayed; although the probability of words (and thus the most probable words) within a topic varies gradually over time, the gist of the topic remains the same.…”

Section: B Hawkes Modelmentioning

confidence: 99%

Mutual-Excitation of Cryptocurrency Market Returns and Social Media Topics

Phillips

Gorse

2018

Proceedings of the 4th International Conference on Frontiers of Educational Technologies

View full text Add to dashboard Cite

Cryptocurrencies have recently experienced a new wave of price volatility and interest; activity within social media communities relating to cryptocurrencies has increased significantly.There is currently limited documented knowledge of factors which could indicate future price movements. This paper aims to decipher relationships between cryptocurrency price changes and topic discussion on social media to provide, among other things, an understanding of which topics are indicative of future price movements. To achieve this a well-known dynamic topic modelling approach is applied to social media communication to retrieve information about the temporal occurrence of various topics. A Hawkes model is then applied to find interactions between topics and cryptocurrency prices. The results show particular topics tend to precede certain types of price movements, for example the discussion of 'risk and investment vs trading' being indicative of price falls, the discussion of 'substantial price movements' being indicative of volatility, and the discussion of 'fundamental cryptocurrency value' by technical communities being indicative of price rises. The knowledge of topic relationships gained here could be built into a real-time system, providing trading or alerting signals.

show abstract

“…The no-arbitrage condition imposes that the excess return µ(t) during a bubble phase is proportional to the crash hazard rate given by Eq. (15). Indeed, setting E[dp] = 0, and assuming that no-crash has yet occurred (dj = 0), this yields µ = κh(t), since E[dj] = h(t)dt by definition of h(t).…”

Section: Appendix B the Log-periodic Power Law Singularity Modelmentioning

confidence: 99%