Interconnection between News Coverage and the Price of Bitcoin in 2017-2019
Galina Borisova
Graduate School of Communication, University of Amsterdam Master Thesis
Dr. Alyt Damstra May 29, 2020
Abstract
This paper studies the bi-directional relationship between financial news and bitcoin price during the most turbulent time in this cryptocurrency’s trading history, namely from 2017 to 2019. I investigate bitcoin related coverage in the two biggest western financial outlets – The Financial Times and The Wall Street Journal – and distinguish between the amount of coverage and the overall sentiment of the articles. Overall, my findings reveal that changes in bitcoin price Granger cause bitcoin related news, especially news with a positive tone, while no support is found for the impact of news on the cryptocurrency’s value. This paper offers new insights into the bitcoin market and its relationship with financial news.
Information plays a crucial role in the occurrence of asset bubbles, such as the recent bitcoin bubble, which lost around 82% of its value in early December 2017 following an earlier skyrocket, a decrease that led to great financial loss among many traders. Nevertheless, even after these billion-dollar losses, bitcoin remains the biggest digital currency in the world with a market capitalization of more than $125 billion (Reiff, 2019) and currently, the price per coin is over $9,000. Such a scale makes that the slightest change in bitcoin price and a good
understanding of the nature of its fluctuations are a very sensitive issue for financial stock traders all around the world.
While bitcoin-related information is mostly transferred via the media, the existent corpus of literature on the relation between media coverage and bitcoin price is very limited ‒ only two academic articles study the impact of social media on bitcoin price (when PhD and Master’s dissertations were not taken into consideration). Philippasa, Rjibaa, Guesmib and Gouttec (2019) find that media attention on social networks, i.e. Twitter and Google Trends, have a partial influence on bitcoin price. Likewise, Mai, Shan, Bai, Wang and Chiang (2018) posit that social media are an important predictor of future values of bitcoin. According to their study, more positive (or negative) posts on an Internet forum (Bitcointalk.org) or Twitter are significantly associated with higher (or lower) next-day bitcoin market prices.
However, there is quite some research that looks into the way in which economic news has a bearing on stock market movements as well as on other economic trends. Several studies
exploring the nature of stock markets have found that media have an effect on it (Boudoukh, Feldman, Kogan, & Richardson, 2013; Kleinnijenhuis, Shcultz, Oegama, & van Atteveld, 2013). Financial news may also affect the real economy. For instance, Wu, Stevenson, Chen and Gner (2002) find that increases in recession-related journalistic pieces in The New York Times
negatively influence real economic changes in the time of worsened economic conditions. Other studies reveal that the tenor of economic news has an impact on consumer confidence (Alsem,
Brakman, Hoogduin & Kuper, 2008; Hollanders & Vliegenthart, 2011), which in turn may have a bearing on the real economy.
The lack of academic research on the relation between media coverage and the price of bitcoin – an asset different from other, more ‘traditional’ stocks – lends academic relevance to the following research question: “How are bitcoin related news coverage and the price of bitcoin related in 2017-2019?”. In addition, the topic is of societal interest as well, given the high stakes that the price of bitcoin has for diverse groups of traders, including the world’s financial elites. With this paper, I aim to investigate the relationship between the amount and tenor of bitcoin-related coverage in The Financial Times and The Wall Street Journal and the price of the virtual currency from January 2017 to December 2019.
Literature review
The specific nature of bitcoin makes it vulnerable to market sentiments. Unlike traditional currencies or other financial instruments that are issued by the central authority (Kristoufek, 2015) and that derive their values from underlying assets (e.g., crude oil, gold), bitcoin is not formally backed by a government or any other legal organization. Its value does not derive from consumption nor from production processes such as, for instance, precious metals (Bouoiyour and Selmi, 2015). All of these peculiarities make bitcoin more reminiscent of the stock than conventional currencies. In this context, it is important to emphasize that the price of bitcoin is solely determined by traders’ operations at the exchange market, as bitcoin trading does not rely on any bank to issue it or anything to be produced to support its price. This makes the bitcoin market a market of the buyer, since there can never be a lack of bitcoin supply. The question remains which factors influence the decision-making process of bitcoin market players and how bitcoin-related information, provided by the media, play a role in this process.
Previously, researchers tried to find the answer to this question by relying on two theoretical frameworks that explain the ground of people’s economic behavior: The Efficient
Market Hypothesis (EMH) and behavioral finance (as a part of behavioral economics). The former postulates that stock prices have a random walk without any underlying logic, which would imply the absence of news effects on market panics (Kleinnijenhuis, Schultz, Oegema, van Atteveldt, 2013). Meanwhile the upholders of behavioral economics have questioned the EMH and argued that financial markets participants often act irrationally governed not solely by practical considerations. Thus, trading decisions could also be shaped and influenced by
emotions, herd, and irrational behavior (Strycharz, Strauss, & Trilling, 2018). In this regard, the media have been identified by scholars to play a significant role in shaping the consensus market opinion and leading the financial ‘herd’ (Oberlechner & Hocking, 2004).
Research has generated robust empirical evidence that EMH is not sufficient to explain the essence of trading (Celeste, Corbet, & Gurdgiev, 2018), whereas the tenants of behavioral finance, to the contrary, were proven to be well applicable in order to understand the nature of online assets (Gurdiev & O’Loughlin, 2020). As a result, there is a growing number of traders who use high-frequency sentiment trading algorithms that embed the evaluation of the news flow directly into their assessment models (Mitra & Mitra, 2011). Put simply, such trading algorithms count the particular words in news articles to identify trends and sentiments related to certain assets. The usage of even such relatively simple technologies ensure a faster reaction and prediction of market fluctuations compared to a situation in which financial decision-makers process the information, discuss it among each other and then change their minds about the value of assets (Mitra & Mitra, 2011). The use of these high-frequency sentiment trading algorithms means that nowadays the news does affect financial markets (Kleinnijenhuis et al., 2013). Within the last decade, there also is a substantial increase of interest in sentiment analysis of news content to enable price predictions among crypto currencies traders (Gurdiev & O’Loughlin, 2020). Despite this growing recognition in the professional trading community of the role that news sentiments play in bitcoin price fluctuations, in academia only a few papers were written about the interconnection between bitcoin price fluctuations and social media, while none of
scholars elaborated on the influence of financial press over bitcoin price. This certainly increases the academic relevance of the current paper.
Thus, the first study by Bouoiyour and Selmi (2015) conducted a quantitative analysis of bitcoin price reactions to investors’ sentiment, that was traced analyzing Google search queries. The authors show that about a fifth of the bitcoin’s price is driven by investor sentiment toward Bitcoin as instrumented by the volume of the Internet users’ inquiries. Kristoufek (2015) looked into the Google and Wikipedia search data for the term ‘Bitcoin’, showing that during the bitcoin bubble which occurred in 2013, the price of this virtual currency was actually led by increased investor interest. Although these studies shed light on the important fact that bitcoin price is – at least partly – driven by traders’ sentiments, the main limitations of these papers are that the Google search inquiries reflect the result of influence over traders’ minds, but not the reason of their economic decisions (e.g. reading of financial news about bitcoin). As mentioned earlier, Mai et al. (2018) as well as Philippas et al. (2019) studied the interconnection between the sentiment of posts in social networks and online forums (i.e. Twitter) and revealed the influence of such information on bitcoin prices. However, these bitcoin-related studies did not attempt to examine the relations between the virtual currency price and the tone of journalistic articles about it.
The lack of research on the interrelationship between financial press coverage and the price of bitcoins is surprising, as there is quite some research focusing on the impact of economic news on the stock market. In fact, the discussion of economic news impact on stock market movements and more broadly on the real economy has been ongoing in academia for decades.
The influence of economic news coverage on public perception of the state of the economy was revealed by Blood and Phillips (1995), whereas Damstra & Boukes (2018) showed that news coverage affects people’s economic expectations but not evaluations. Wu et al. (2002) inferred that the increase of recession-related journalistic pieces in The New York Times influences real economic changes in the time of worsened economic conditions. Other studies
revealed that the tone of economic news affects consumer confidence (Hollanders & Vliegenthart, 2011) that in its turn impairs the state of the real economy.
Meanwhile several studies that explored the nature of stock markets found an effect of social media on it (Boudoukh, Feldman, Kogan, & Richardson, 2013; Kleinnijenhuis, Shcultz,
Oegama, & van Atteveld, 2013). One of the key findings in this area is that the more media attention is given to a specific stock or share the higher the trading volumes are (e.g. Boudoukh, 2012). Thus, Engelberg and Parsons (2011) analyzing the simultaneous reactions of investors in 19 local markets to the same set of information events (earnings releases of S&P 500 Index firms) found that the presence or absence of local media coverage is strongly related to the probability and magnitude of local trading. In other words, the more the audience is exposed to information about a certain asset, the more important it becomes in their minds. This observation is well in line with the idea of agenda-setting, which assumes that media attention for a certain topic increases the salience of this same topic in the minds of people, setting it higher on the public agenda (Carroll & McCombs, 2003). Based on these remarks I expect the following:
H1: Change in the amount of bitcoin news leads to change in the price of bitcoin.
In addition, work points to a positive relation between the tone of media coverage and the stock prices. Tetlock (2007) revealed that high levels of pessimism in economic news in The Wall Street Journal robustly predict downward pressure on market prices. Likewise, Soroka, Stecula, and Wlezien (2015) inferred that the tone of economic articles in The New York Times and The Washington Post had a positive relation with economic evaluations of its readers. In line with this, Strycharz et al. (2018) infered that increases in positive news about ING, Philips and Shell led to the rise in prices of these companies’ shares. These theoretical considerations allow me to hypothesize that:
H2: The tone in bitcoin news is positively related to the price of bitcoin, meaning that a
positive tone leads to an increase in bitcoin price and a negative tone leads to a decrease in bitcoin price.
With regard to the audience reaction to the news sentiments some researchers (Van Dalen, de Vreese, & Albæk, 2017; Damstra, Boukes, & Vliegenthart, 2020) suggest that people respond more strongly to negative economic information than to positive economic information. Such asymmetric responsiveness could be explained by the fact that people care more strongly about a potential loss than they do about a gain of the same magnitude (Tversky & Kahneman, 1974).
This negativity effect has been studied extensively across research fields and is also empirically confirmed in work on economic news effects: Negative economic news makes people more pessimistic about the national economy, whereas positive economic news does not provoke the equivalent countereffect (e.g., Damstra & Boukes, 2018; Soroka, 2006). Thus, Damstra et al. (2020) while studying the effects of credit and blame attributions in economic news on government evaluations revealed a certain asymmetry in audience responsiveness to political media coverage. Specifically, when the government is blamed by the media for a worsened economic situation, the consequences are real: news consumers become more likely to also accuse politicians of causing the crisis, and, partly as a result of that, they evaluate the government’s economic performances more negatively. However, when the government is praised for the economic success in the news, the opposite causal chain cannot be applied.
According to Strauss (2018), more negative emotional words used in articles dealing with specific stocks leads to a decrease of the opening prices of those stocks. The researcher
demonstrated how the positive tone of news has only a moderate positive impact on the price, while negativity in the articles is a rather strong predicter of drops in the opening price of stocks the following day. Based on this, I expect the following:
H3: The negative impact of negative tone on bitcoin price is stronger than the positive
impact of positive tone.
In addition to these media effects, one could also anticipate that price fluctuations at the stock market co-determine what kind of news will be published. In their study on the influence of emotions in Dutch newspapers over the stock prices of 21 companies, Strauss, Vliegenthart, and Verhoeven (2016) revealed that change in stock prices Granger causes the intensity of media coverage of it. This implies that financial daily media coverage could not only shape stock markets, but also in fact mirror them.
Other scholars emphasize that journalists cover economic developments rather like a burglar alarm than a police patrol (Zaller, 2003), drawing attention to significant developments only. In other words, if nothing extraordinary happens in the economy, media coverage of it is fairly routine. When something appears to be wrong, the economy demands front page (Goidel & Langley, 1995). Therefore, economic news reflects change in economic developments rather than the absolute state of the economy (Van Dalen et al., 2017). This assertion is also supported by new values theory (Galtung & Ruge, 1965). The theory implies that there are several
attributes of stories (e.g., proximity, novelty) that determine whether certain news or topics are selected to be covered by media. Thereby, it can be suggested that market movements that bring novelty (e.g., skyrocket or slump in assets’ value) are more likely to be covered by journalists than more regular market developments. Based on this claim I anticipate:
H4. Change in the price of bitcoin leads to change in the amount of coverage
Like the negative tone of the news could have a greater impact on the bitcoin price, the slump in the virtual currency price could draw more attention from journalists than a similar increase would do. Some earlier studies have demonstrated how news becomes more negative when the economy worsens, but not more positive when the economy improves (Blood & Philips 1995;
Goidel & Langley 1995). Public responses to negative economic information tend to be greater than public responses to positive economic information. The same trend is evident in mass media content, and this content serves to enhance the asymmetry in public responsiveness (Soroka, 2006). Mass media respond asymmetrically to economic information, and the public then responds asymmetrically to both media content and to the economy itself.
Furthermore, according the Van Dalen et al. (2017) study of the relation between economic developments and the tone and visibility of economic news in Danish newspapers, journalists only react to negative changes and not to positive ones. Researchers explain this observation from the perspective of the surveillance function of economic news, since negative developments require the awareness from the rather inattentive audience. Although economic news become more positive in tone during the boom period, it does not become more visible. When the economic media coverage boosts the negative state of the economy, on the other hand, it also becomes more visible, as the researchers revealed. Therefore, I formulate the following hypothesis:
H5: The effect of negative change in bitcoin price on media coverage is stronger than the
effect of positive change
Drawing on the literature discussed above I intend to use the following conceptual definition as the basis for the current paper, the way in which the concept is operationalized is discussed in the methods section.
Influence of the media on bitcoin price ‒ The influence of media coverage on price asset, which runs through the decision making process of financial actors (Donaldson & Preston, 1995) and can be explained by the fact that traders change their economic decisions based on their sentiments toward the asset (Strycharz et al., 2018). Such sentiments are influenced by the nature of media coverage (positive/negative).
Data and Method
To answer my research question, I collected bitcoin-related news coverage published on the websites of the two biggest financial media outlets: The Financial Times and The Wall Street Journal. These newspapers have the biggest circulation figures among financial outlets in the world (Robins, 2018), and exert considerable influences on their audiences. Both newspapers publish in English and can be easily accessed online, which adds to their frequent use by the international online trading community as English is the most commonly used language in the world.
To find the articles the key word ‘bitcoin’ is used in search engines of both websites. The time frame of the research is from the 1st of January 2017 to the 31st of December 2019 since the most unprecedented bitcoin price fluctuations occurred in this timespan.1 The focus of my
research is on news in which the bitcoin phenomenon is discussed in a substantive way. To ensure the sample deals with this type of data, the selected news articles were manually
scrutinized and all items in which only very brief and superficial references to bitcoin were made were excluded, as were press releases, caricatures (that were published in the Financial Times), videos and graphics. The final dataset comprises of 1,130 news articles, 704 of them published on the Financial Times website and 426 appeared on the website of The Wall Street Journal. Operationalization
Studies that examine media influence on financial markets, often adopt a two-step approach. First, researchers conduct a content analysis of news discussing certain assets. Second, the statistical analyses are carried out in order to find out whether the amount and tone of articles have a correlation with the analyzed assets’ (e.g. companies’ stocks) price move or public economic perception (Engelberg & Parsons, 2011; Kleinnijenhuis et al., 2015; Strycharz et al., 2018; Strauss, 2018; Damstra & Boukes, 2018 etc.). By doing so, academics have aimed to
reveal whether and to what extent an increase in the number of (positive) articles leads to rises in the assets’ value and vice versa.
For this thesis, I adopt a similar approach and for the operationalization of my key
concepts I try to build on validated examples from the literature, as much as possible. Seven time series are operationalized as the variables for the further analysis, namely, number of articles about bitcoin, positive and negative tone of bitcoin-related coverage, bitcoin price, negative and positive change in bitcoin price and the EU unemployment rate. All the values are measured on a weekly basis and further explained below.
Tone of the news
To identify positive and negative sentiment in bitcoin related news, I rely on the search terms developed by Kleinnijenhuis et al. (2015) to measure sentiment in financial news and later also employed by Damstra and Boukes (2018) to study tone in economic news coverage. The complete list of words to identify the news sentiment for creating codebook and conducting an analysis was obtained. According to the mentioned approach, ‘positive tone has been measured by counting the number of references to hope, confidence, enthusiasm, inspiration, relief, grip, rescue, and recovery(-related) words. Negative tone has been measured by counting the
references to fear, shock, panic, danger, worry, disturbance, stress, tension, and anxiety(-related) words’ (Damstra & Boukes, 2018). To verify this measurement’s validity as well as to extend the list of search terms I manually analyzed 50 randomly selected articles from the sample. As a result, the list of words for identifying the tone of the articles was extended to 644 words list (see the search terms in Appendix). The new extended list is believed to boost the validity of the current study, because it is better adjusted to the selected financial media, that certainly have some peculiarities of the used vocabulary.
Following the practice of some researchers that carried out automated content analysis (Strauss et al., 2016; Strycharz et al., 2018) I wrote a code using Python programming language and embedded the extended list of search terms in it (see the code in Appendix). The code finds,
counts and summarizes the number of positive and negative references per article, and it calculates the overall tone of the articles by deducting the number of negative references from the number of positive ones. Then it splits the overall sentiment of the articles in two groups in the following way: if the overall tone is less than zero (has a negative sign) the article is
categorized as negative, if the value of overall tone is higher than zero, the article is categorized as positive. The values of the number of articles as well as the positive tone and negative tone variables were converted into the mean of the weekly number of articles using Python.
Therefore, the ‘positive tone’ variable indicates the mean of overall positive articles published in both news websites and likewise for the ‘negative tone’ variable.
Table 1. Descriptive statistics.
N Minimum Maximum M SD Number of articles 156 1 55 7.244 7.751 Positive tone 156 0 27 3.538 4.087 Negative tone 156 0 25 3.282 4.062 Bitcoin price, $ 156 821.8 19140.8 6348.15 3534.01 Positive Δ in BTC price, $ 156 0 4132.2 332.996 634.524 Negative Δ in BTC price, $ 156 0 5215 294.917 684.82 Unemployment rate, % 156 6.2 8.1 6.939 0.579
Note. N indicates the number of observations (weeks)
Bitcoin Price
Bitcoin daily price history was retrieved from Yahoo Finance (this source was also used by Vliegenthart and Damstra (2019) in their study of the interactions among parliamentary questions, newspaper coverage on the economic crisis, and consumer confidence) and converted to the weekly level (mean value per week) using Python. The positive and negative changes in bitcoin price were created as two variables to help to answer whether the effect of negative change in bitcoin price on article sentiment is stronger than positive change (Hypothesis 5). For
change, all values were converted into absolute (positive) values regardless of the direction, as was done by Damstra and Boukes (2018). This eases the interpretation and implies that positive scores (in case of negative change) implies negative developments (Table 1).
Unemployment Rate
I control for real-world developments by adding the unemployment rate to the equation. This measure serves as an indicator of general economic developments, which also may be of influence on traders’ decision to buy or sell bitcoins.
Analytical procedure
After getting the results of the automated content analyses I conducted two different tests. First, I looked into the correlation between the variables as it was performed by Vliegenthart and
Damstra (2019) in their study of the interactions among parliamentary questions, newspaper coverage on the economic crisis, and consumer confidence.
Because the causal direction of the relationships between the variables cannot be determined beforehand, I adopt Vector Autoregression (VAR) modelling to assess how each series may have an impact on the other ones, as it was employed by several researchers in media studies (Vliegenthart, 2014; Strauss et al., 2016; Strycharz et al., 2018). This analysis is
particularly appropriate for my research setup as it tests the interdependence of the variables. In this sense, both the media variables and the bitcoin price variables are considered as dependent and independent variables at the same time in the model (Strycharz et al., 2018). Given that all variables used in the analysis are time series, I assess whether changes in lagged values of one variable lead to changes in current values of the other variables, an approach that allows for the presence of bi-directional relationships (Vliegenthart & Damstra, 2019).
Before the analyses can be run, it is necessary to check whether the series are stationary or not. To do so, I conduct Dickey Fuller tests, as recommended by Vliegenthart (2014). The bitcoin price and the unemployment rate variables turn out to be non-stationary, while the rest of
the variables are stationary. Therefore, I differenced the variables, which implies that I look at changes instead of at levels in the interpretation of the results. Then I select the optimal number of lags looking at a maximum of 20 lags, a one-lagged model and a three-lagged model appear to be most appropriate. I chose to build the one-lagged model, because I expect media effects on the price of bitcoin (and vice versa) to occur in relative short time frames (probably even shorter than a week, I measured the variables in the study on a weekly level). Granger-causality tests were carried out to examine whether one series predicts another series above and beyond the past values of its own series. I also performed several tests to eliminate common estimation errors. The Portmanteau (Q) test was used to rule out serial correlation of the residuals up to the lag order of 20. Next, the Portmanteau (Q) test was applied to the squared residuals of the series to check for heteroscedasticity. The test showed that three of the residuals move together to a high degree (0.77-0.90). This could imply that a lower aggregation level would be more optimal for the analysis of the variables. However, while collecting the articles I noticed that bitcoin-related articles often were not published on a daily basis, which makes it difficult to change the
measurement level. Given the space restrictions, I only report the significant results of the analysis.
Results
Before looking into causal relationships, Table 2 shows the correlation between number of articles published about bitcoin, number of overall positive and negative publications, bitcoin price and its positive and negative change and unemployment rate. Most of the mentioned variables have moderate and even strong associations.
Table 2. Correlation coefficients between all variables
Variable Number of
articles
BTC price Positive tone Negative tone
Number of articles 1.000
Positive tone .901* .472* 1.000
Negative tone .896* .431* .624* 1.000
Unemployment rate .131 -.501* .106 .143
Note. N=156 (number of weeks). BTC price=bitcoin price.
There is a strong significant positive correlation (r = .503) between bitcoin price and number of articles about bitcoin. Figure 1 depicts both trends over time. Throughout the course of 2017, bitcoin price increased along with the amount of journalistic stories published in the financial outlets, from January 2019 onwards, both indicators went gradually down again.
Figure 1. Volume of the news about bitcoin and bitcoin price curve.
With regard to the tone of articles positive publications have a moderate positive
association with bitcoin price (r = .47) meaning that with the increase in bitcoin price amount of positive journalistic pieces rises. This tendency was somewhat stronger at the end of 2017 – beginning of 2018, according to Figure 2.
Figure 2. Number of positive articles and bitcoin price.
Somewhat surprisingly, there is a positive correlation between the virtual currency’s values and the number of negative articles (r = .43), according to Table 2. In other words, increases in the price of bitcoin is related to more positive as well as more negative news coverage, which runs counter to my expectation as formulated in Hypothesis 2. However, to formally test Hypothesis 2 I relied on VAR analyses and Granger causality tests to see whether a positive tone leads to an increase in bitcoin price and a negative one to the opposite.
Media coverage and the price of bitcoin
Table 3 presents the results of the VAR-analysis, on the basis of which I am able to answer my research question as well as hypotheses. Table 4 displays the direction of relationships and presents how the variables influence one another. The VAR-table shows that changes in the number of articles do not lead to changes in bitcoin price (the correlation between variables is insignificant p = .894). Also, Table 4 demonstrates that number of articles is not Granger causing bitcoin price. Therefore, Hypothesis 1 is not supported.
The results of the VAR-analysis and the Granger causality tests (Table 3 and Table 4 respectively) indicate that there is no significant positive relationship between the tone of the news and change in bitcoin price, meaning that a positive tone does not lead to an increase in bitcoin price and a negative tone does not lead to a decrease in its value. Since no statistically significant relationships between these variables were found I must reject Hypothesis 2.
Table 3. Relations between articles and bitcoin price.
Independent variables
Dependent variables Number of
articles
BTC price Positive tone Negative tone
Number of articles -.314 (.465) 12.315 (92.165) .143 (.269) .016 (.292) BTC price .002** (.001) -.107 (.120) .001** (.000) .001 (.000) Positive tone -.0136 (.499) -26.519 (98.846) -.500 (.288) -.001 (.314) Negative tone .467 (.506) 9.786 (100.192) .136 (.292) -.203 (.318) Unemployment rate 2.711 (12.636) 375.329 (2504.054) 7.374 (7.295) -5.341 (7.947) Constant -.057 (.449) 52.016 (88.894) .056 (.259) -.113 (.282) R-squared .141 .020 .297 .118
Note. N=154 (number of weeks). BTC price=bitcoin price. All variables are differenced. Standard errors are indicated in the parentheses.
*p<.05. **p<.01. ***p<.001
Likewise, since no statistically significant association between the tone of articles and bitcoin price were found I cannot prove that the negative impact of negative tone is stronger than the positive impact of positive tone. Based on this assertion I reject Hypothesis 3.
While scrutinizing the impact of changes in the value of the online currency on changes in news coverage, I found statistically significant relationships. Namely, changes in bitcoin price are significantly related to changes in the number of articles (Table 3); changes in bitcoin price are found to be Granger causing changes in the volume of coverage, leasing to increasing numbers of publications (Table 4). Based on this, I accept Hypothesis 4.
Table 4. Granger causality test results, Prob > chi2*.
Independent variables Dependent variables
Number of articles BTC price Positive tone Negative tone Positive Δ BTC price Negative Δ BTC price Number of articles ‒ .894 .594 .955 .346 .478 BTC price .002* ‒ .001* .115 .000* .000* Positive tone .978 .788 ‒ .997 .411 .724 Negative tone .355 .922 .642 ‒ .652 .604 Positive Δ BTC price .731 .537 .071 .060 ‒ .001* Negative Δ BTC price .419 .300 .187 .017 .000* ‒ Unemployment rate .830 .881 .312 .501 .848 .657
Note. The variables in the rows present the IVs and the variables in the columns serve as DVs. *the result is significant if the p-value <.05
Additional analyses were conducted to assess whether the impacts of positive and
negative change in the price of bitcoin provoked different effects on coverage. The VAR analysis results show that there is no significant effect of positive change in bitcoin price, whereas the negative change in the currency’s value leads to a decrease in the amount of negative
publications on a weekly basis (p = .017). It is important to clarify, that he fact that I used the absolute values of negative change in bitcoin price (having converted them into positive scores) does not change the direction of the effect. This finding certainly contradicts my initial
assumption and allows me to reject Hypothesis 5.
In addition, the results showed that changes in bitcoin price have a positive influence on changes in the number of positive journalistic pieces about it (Table 3); the lagged change in bitcoin price Granger causes increases in positive news (Table 4).
Consequently, I can answer this study’s overarching research question “How are bitcoin related news coverage and the price of bitcoin related?” in the following way: Lagged changes in bitcoin price Granger cause current changes in the number of news articles about bitcoin as well as an increase in positive coverage published in The Financial Times and The Wall Street Journal. Controversially, negative change in bitcoin price preceded a decrease in the number of negative publications.
Discussion
With this study, my aim was to investigate the interrelationship between the price of bitcoin and bitcoin-related news coverage in the two biggest financial newspapers during the most dramatic period in bitcoin price history ‒ 2017-2019. Specifically, I wanted to find out whether the
amount of news and the sentiment of the news had an influence over the virtual currency price and vice versa.
The results of the VAR analyses partly confirm some findings from earlier research in adjacent fields, namely, the influence of financial news on stock market movements. For
instance, my analyses confirm that change in bitcoin price Granger cause an increase in number of news articles about bitcoin in the next week. This is fully in line with findings by Strrauss et al. (2016) and Van Dalen et al. (2017), who claim that financial news in times of economic distress tends to be more intensive compared to periods of stabilization. This could be explained by referring to news values theory, that states how change and novelties are generally considered newsworthy (Galtung & Ruge, 1965). That said, my analyses did not provide empirical support for another tenet of news values theory, namely that negative trends evoke more coverage than positive ones. In other words, the decline in bitcoin price does not have a greater influence on media coverage than a rise in the virtual currency value does.
Moreover, neither the number of articles in the financial press nor the positive or negative tone of these journalistic pieces were proven to have a statistically significant impact on bitcoin price. This revelation certainly contradicts findings of other studies, that claims the opposite with regard to the stock prices and economic evaluations of the general public (Strycharz et al., 2018; Damstra & Boukes, 2018). The result that the change in bitcoin price Granger causes increases in positive coverage does not resemble previous results and it might be interesting to further
investigate this effect in the future.
Overall, my research shows that, as it was discussed in the introduction and literature review sections, bitcoin differs from any offline currencies and virtual assets. Since bitcoin has no underlying asset or economic indicators behind its value, in contrast to most other stocks, where a company's earnings and profitability serve as fundamental factors that move their price2.
The findings of this paper are in line with the results of Strauss et al. (2016), who revealed that the sentiments of the studied articles do not show the consistent effect on the opening prices of the stock the following day. According to scholars’ analysis, newspapers rather reflect
movements on the stock market, that predict it, which was also supported by the finding of the current paper.
Since no statistically significant predictor of bitcoin price was found, it could be an interesting avenue for future research to conduct a panel survey among bitcoin traders to identify the factors that navigate their trading decisions, including, the kind of information that is
available in the public domain that has an influence on their behavior at the exchange. Inevitably, my paper faces some limitations, including the chosen research method. Although computer-based content analysis has the great advantage of relatively high reliability (the measurement is consistent and can easily be reproduced in another research setup), the internal validity could be threatened. Given that this type of analysis is not the most sensitive measure when it comes to nuanced differences in meaning as well as indirect/hidden semantics (e.g. irony, sarcasm) some discrepancies could appear.
Overall, amid the lack of academic literature on the interconnection between media coverage and bitcoin price the current paper is a first attempt to shed a light on this undoubtedly complicated relationship. By doing so, I hope to have provided some venues for future academic investigation of this unexplored topic.
Bibliography
Alsem, K. J., Brakman, S., Hoogduin, L., & Kuper, G. (2008). The impact of newspapers on consumer confidence: does spin bias exist? Applied Economics, 40(5), 531-539. Bitcoin price history. (2020, May 26). Retrieved from Yahoo Finance:
https://finance.yahoo.com/quote/BTC-USD/history?period1=1483315200&period2=1577750400&interval=1wk&filter=history &frequency=1wk
Blood, D. J., & Phillips, P. C. (1995). Recession headline news, consumer sentiment, the state of the economy and presidential popularity: A time series analysis 1989–1993. International Journal of Public Opinion Research, 7(1), 2-22.
Boudoukh, J., Feldman, R., Kogan, S., & Richardson, M. (2013). Which news moves stock prices? A textual analysis (No. w18725). National Bureau of Economic Research.
Bouoiyour, J., Selmi, R. (2015). What does Bitcoin look like? Annals of Economics and Finance, 16(2), 449-492.
Carroll, C. E., & McCombs, M. (2003). Agenda-setting effects of business news on the public's images and opinions about major corporations. Corporate reputation review, 6(1), 36-46. Celeste, V., Corbet, S., & Gurdgiev, C. (2018). Fractal Dynamics and Wavelet Analysis: Deep
Volatility Properties of Bitcoin, Ethereum and Ripple.
Damstra, A., & Boukes, M. . (2018). The Economy, the News, and the Public: A Longitudinal Study of the Impact of Economic News on Economic Evaluations and Expectations. Communication Research, 1-25.
Damstra, A., Boukes, M., & Vliegenthart, R. (2020). To Credit or to Blame? The Asymmetric Impact of Government Responsibility in Economic News. International Journal of Public Opinion Research.
Damstra, A.,Boukes, M., Vliegenthart, R. (2018). The economy. How do the media cover it and what are the effects? A literature review. Sociology Compass, 12(5).
Engelberg, J. E., & Parsons, C. A. (2011). The causal impact of media in financial markets. The Journal of Finance, 66(1), 67-97.
European Union unemployment rate. (2020, May 26). Retrieved from Trading Economics: https://tradingeconomics.com/european-union/unemployment-rate
Galtung, J., & Ruge, M. H. (1965). The structure of foreign news: The presentation of the Congo, Cuba and Cyprus crises in four Norwegian newspapers. Journal of peace research, 2(1), 64-90.
Goidel, R. K., & Langley, R. E. (1995). Media coverage of the economy and aggregate economic evaluations: Uncovering evidence of indirect media effects. Political Research Quarterly, 48(2), 313-328.
Gurdgiev, C., & O’Loughlin, D. (2020). Herding and anchoring in cryptocurrency markets: Investor reaction to fear and uncertainty. Journal of Behavioral and Experimental Finance, 25, 100271.
Hollanders, D., & Vliegenthart, R. (2011). The influence of negative newspaper coverage on consumer confidence: The Dutch case. Journal of Economic Psychology, 32(3), 367-373. Kleinnijenhuis, J., Schultz, F., Oegema, D, van Atteveldt, W. (2013). Financial news and market
panics in the age of high-frequency sentiment trading algorithms. Journalism, 14(2) , 271–291.
Kristoufek, L. (2015). What are the main drivers of the Bitcoin price? Evidence from wavelet coherence analysis. PloS one, 10(4).
Mai, F., Shan, Z., Bai, Q., Wang, X., Chiang, R.H.L. (2018). How Does Social Media Impact Bitcoin Value? A Test of the Silent Majority Hypothesis. Journal of Management Information Systems, 35 (1), 19–52.
Mitra, G., Mitra, L. (2011). The handbook of news analytics in finance (Vol. 596). John Wiley & Sons.
Oberlechner, T., & Hocking, S. (2004). Information sources, news, and rumors in financial markets: Insights into the foreign exchange market. Journal of economic psychology, 25(3), 407-424.
Philippasa, D., Rjibaa, H., Guesmib, K., Gouttec, S. (2019). Media attention and Bitcoin prices. Finance Research Letters 30 , 37–43.
Reiff, N. (2019, June 25). Top 5 Cryptocurrencies by Market Cap. Retrieved from Investopedia: https://www.investopedia.com/news/top-5-cryptocurrencies-market-cap/
Soroka, S. (2006). Good News and Bad News: Asymmetric Responses to Economic Information. The Journal of Politics, 68(2), 372-385.
Soroka, S. N., Stecula, D. A., & Wlezien, C. (2015). It's (change in) the (future) economy,
stupid: economic indicators, the media, and public opinion. American Journal of Political Science, 59(2), 457-474.
Strauß, N. (2018). News media and the stock market: Assessing mutual relationships: An interdisciplinary multimethod.
Strauß, N., Vliegenthart, R., & Verhoeven, P. (2016). Lagging behind? Emotions in newspaper articles and stock market prices in the Netherlands. Public Relations Review, 42(4), 548-555.
Strycharz, J., Strauss, N., & Trilling, D. (2018). The role of media coverage in explaining stock market fluctuations: Insights for strategic financial communication. International Journal of Strategic Communication, 12(1), 67-85.
Tetlock, P. C. (2007). Giving content to investor sentiment: The role of media in the stock market. The Journal of finance, 62(3), 1139-1168.
Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124-1131.
Van Dalen, A., de Vreese, C., & Albæk, E. (2017). Economic news through the magnifying glass: How the media cover economic boom and bust. Journalism Studies, 18(7), 890-909.
Vliegenthart, R., & Damstra, A. (2019). Parliamentary questions, newspaper coverage, and consumer confidence in times of crisis: A cross-national comparison. Political Communication, 36(1), 17-35.
Vliegenthart, R., & Mena Montes, N. (2014). How political and media system characteristics moderate interactions between newspapers and parliaments: Economic crisis attention in Spain and the Netherlands. The International Journal of Press/Politics, 19(3).
What influence a share price. (2020, May 26). Retrieved from London Stock Exchange: https://www.londonstockexchange.com/traders-and-brokers/private-investors/investing-in-shares/about-share/what-influence-share-price/what-influence-share-price.htm Wu, H., Stevenson, R., Chen, H., & Gner, Z. (2002). The Conditioned Impact of Recession
News: A TimeSeries Analysis of Economic Communication in the United States, 19871996. International Journal of Public Opinion Research, 14(1), 19-36.
Zaller, J. (2003). A new standard of news quality: Burglar alarms for the monitorial citizen. Political Communication, 20(2), 109-130.
Appendix
Code 1. Python code, that was used for the analysis.
import pandas as pd import seaborn as sns pd.set_option('display.float_format', lambda x: '%.3f' % x) %matplotlib inline import datetime import numpy as np sample=pd.read_csv('Final sample.csv')
Finding and counting the references.
def word_frequency(text, query): import re
text = str(text).lower() newquery = [] for word in query:
newquery.append(str(word).lower()) tokens = re.findall(r"[\w']+|[.,!?;$@#]", text) counter = 0
for word in tokens: if word in newquery: counter += 1 return counter
Embedding the search terms for positive and negative references.
sample['positive'] = sample['Text'].apply(word_frequency,
args=(['hope', 'hopes', 'hopped', 'look forward', 'looking forward', 'looks forward', 'dominate', ' dominates', 'ICO-friendly', 'resolve', 'resolves', 'resolved', 'envision', 'growing', 'grow', 'grows', 'grew', 'successful', 'su ccessfully', 'professional', 'professionally', 'uptrend', 'trust', 'trusts', 'trusted', 'secure', 'secured', 'secures', 'advantage', ' advantages', 'backers', 'familiarity', 'reliability', 'reliable', 'comfort', 'comfortable', 'valued', 'wealthiest', 'devout', 'belie vers', 'believer', 'support', 'supporters', 'supporter', 'decent', 'upside', 'powerful', 'promise', 'promises', 'promised', 'pro mising', 'support', 'supports', 'supported', 'supporting', 'inspire', 'inspires', 'inspiring', 'inspired', 'inspiration', 'earn', 'ear
ns', 'earned', 'recover', 'recovered', 'recovers', 'recovering', 'reassure', 'excellent', 'excellence', 'reassures', 'fabulous', 're assured', 'reassuring', 'reassurance', 'generous', 'assurances', 'assurance', 'assure', 'assured', 'save', 'good', 'better', 'nice', 'best', 'safe', 'saved', 'safety', 'saves', 'happy', 'favourable', 'favourite', 'abuzz', 'hopeful', 'exultations', 'exultation', 'stabl e', 'stability', 'stabilize', 'stabilized', 'security', 'growth', 'peaked', 'peak', 'rise', 'rose', 'risen', 'rises', 'soar', 'soared', 'soar s', 'high', 'highs', 'higher', 'highest', 'picked', 'pick', 'picks', 'relieve', 'relieved', 'acceptance', 'accepted', 'accept', 'accept able', 'accepts', 'accepting', 'safeguards', 'easy', 'ease', 'easier', 'increase', 'increased', 'increases', 'increasing', 'confidenc e', 'confident', 'supportive', 'advanced', 'valuable', 'cognisant', 'jump', 'jumps', 'jumped', 'jumping', 'protections', 'protec t', 'protects', 'protection', 'faith', 'effective', 'effectiveness', 'passion', 'passionate', 'afficionados', 'afficionado', 'fairness' , 'integrity', 'security', 'flowed', 'smoothly', 'smooth', 'best-known', 'reach', 'reached', 'climb', 'climbs', 'climbed', 'buoye d', 'perspective', 'perspectives', 'visible', 'visibility', 'clear', 'transparency', 'transparent', 'satisfactorily', 'satisfaction', 'h elp', 'helping', 'helps', 'lifted', 'lifts', 'lift', 'welcoming', 'wiser', 'wise', 'wisdom', 'bullish', 'optimists', 'optimism', 'prove n', 'spike', 'refuge', 'refuges', 'interest', 'interested', 'interesting', 'believe', 'believes', 'belief', 'raise', 'upward' 'gain', 'gai ns', 'gaining', 'gained', 'surge', 'surges', 'surging', 'surged', 'safe', 'safer', 'save', 'saved', 'saving', 'saves', 'excitement', 'pr omise', 'promises', 'promised', 'exciting', 'excitement', 'benefits', 'benefit', 'benefitted', 'attractive', 'attracts', 'attractions ', 'attract', 'attracting', 'attracted', 'enthusiasts', 'enthusiast', 'rebound', 'rebounded', 'rebounding', 'rebounds', 'enthusias m', 'enthusiastic', 'positive', 'win', 'won', 'winner', 'winning', 'legally', 'prize', 'prosperous', 'prosperity', 'legal', 'solidifyi ng', 'solid', 'popular', 'strength', 'strong', 'stronger', 'strongest', 'bull', 'strengthened', 'reasonable', 'insurance', 'legitimat e', 'rocketed', 'rocket', 'rockets', 'prestige', 'crypto-euphoria'],))
sample['negative'] = sample['Text'].apply(word_frequency,
args=(['fear', 'fears', 'feared', 'fearful', 'falsely', 'false', 'bots', 'bot', 'warning', 'warn', 'reduced', 'r educed', 'reducing', 'scare', 'scared', 'scares', 'wreck', 'wrecked', 'wrecks', 'shock', 'shocked', 'shocks', 'lack', 'vulnerable' , 'dim', 'dims', 'attack', 'attacks', 'terrible', 'terribly', 'arrested', 'arrests', 'arrest', 'unlicensed', 'dip', 'dipping', 'dipped', 'un authorized', 'shutdown', 'suffer', 'suffering', 'suffered', 'suffers', 'conspicuously', 'startle', 'victim', 'victims', 'disadvanta ge', 'disadvantages', 'hindrance', 'pressure', 'shadow', 'panic', 'panics', 'unlawfully', 'pitfalls', 'pitfall', 'scepticism', 'abus ive', 'worry', 'worried', 'worries', 'unclear', 'bleak', 'distraction', 'distracted', 'distracts', 'depress', 'depressed', 'depressiv e', 'desperate', 'desperate', 'opaqueness', 'drop', 'dropped', 'drops', 'tumbled', 'tumble', 'tumbles', 'decline', 'declined', 'de clines', 'war', 'shut', 'disruption', 'disruptive', 'odd', 'steal', 'steals', 'stole', 'stollen', 'disrepute', 'derail', 'derails', 'derailed ', 'controversial', 'damaged', 'unpaid', 'crazy', 'uncertainties', 'crash', 'crashing', 'crashed', 'down', 'collapse', 'collapsed', 'collapses', 'abnormal', 'ban', 'banned', 'skew', 'skewed', 'fads', 'detriment', 'harm', 'harms', 'fanned', 'mis-selling', 'tensi on', 'ransomware', 'guilt', 'guilty', 'embezzlement', 'embezzle', 'embezzled', 'embezzles', 'anxiety', 'anxious', 'bubble', ' bubbles', 'vulnerable', 'criticisms', 'riskier', 'plunge', 'nervousness', 'low', 'lows', 'lowest', 'red-handed', 'crook', 'cumber
some', 'whizzy', 'unrest', 'difficult', 'difficulties', 'scams', 'scam', 'scammed', 'manipulate', 'manipulates', 'manipulation' , 'manipulator', 'manipulators', 'sued', 'sue', 'sues', 'weakest', 'weak', 'downward', 'dumb', 'idiot', 'difficulty', 'difficultie s', 'skeptics', 'dismissive', 'overconfident', 'awfully', 'awful', 'turmoil', 'busted', 'bust', 'busts', 'dark', 'secretly', 'scandals' , 'crises', 'crisis', 'slides back', 'slide back', 'gave up', 'give up', 'gives up', 'struggle', 'struggles', 'struggling', 'struggled', 'anger', 'angry', 'fell', 'fall', 'fallen', 'falls', 'danger', 'dangers', 'dangerous', 'frenzy', 'scandal', 'violent', 'defaults', 'default ', 'perils', 'peril', 'losing', 'concerns', 'concern', 'concerned', 'concerning', 'bankrupt', 'bankrupts', 'lose', 'lost', 'loss', 'loss es', 'loser', 'risk', 'risks', 'risky', 'sceptical', 'mania', 'drawbacks', 'drawback', 'illegal', 'illegally', 'fraud', 'hacking', 'hack' , 'hacked', 'hacks', 'hackers', 'hacker', 'speculation', 'speculative', 'speculations', 'craze', 'threat', 'threats', 'threatening', ' threatened', 'worse', 'threaten', 'suspicion', 'suspicious', 'suspiciously', 'crazy', 'challenge', 'problem', 'problems', 'challe nges', 'crime', 'crimes', 'criminal', 'criminality', 'criminals', 'fraudsters', 'fraudster', 'negative', 'irresponsible', 'fraudulen t', 'crackdown', 'clampdown', 'penalties', 'penalty', 'overheated', 'misconduct', 'violation', 'laundering', 'terrorist', 'terror ism', 'terrorists', 'shut', 'shuts', 'regret', 'regrets', 'regretted', 'bear', 'plummeted', 'plummet', 'plummets', 'theft', 'outlaw', 'antipathy', 'dent', 'dented', 'shutting', 'tamp', 'loosely', 'poor', 'poorly', 'robbed', 'robbery', 'rob', 'robs', 'mugger', 'robber ', 'conflicts', 'conflict', 'conflicted', 'failing', 'fail', 'failure', 'fad', 'smuggler', 'launders', 'launder', 'drug', 'drugs', 'hurting' , 'hurt', 'hurts', 'inconvenience', 'inconvenient', 'obnoxious', 'manias', 'unease', 'stolen', 'dismissed', 'stupid', 'confiscate d', 'confiscate', 'confiscates'],))
Computing overall sentiment and overall positive and overall negative variables. sample['overall_sentiment'] = sample['positive'] - sample['negative']
sample['overall_negative'] = sample['overall_sentiment'] < 0 sample['overall_positive'] = sample['overall_sentiment'] > 0 sample['overall_neutral'] = sample['overall_sentiment'] == 0 sample['overall_negative']=sample.overall_negative.astype(int) sample['overall_positive']=sample.overall_positive.astype(int) sample['overall_neutral']=sample.overall_neutral.astype(int)
Converting data to the weekly level.
from datetime import datetime
sample['Date '] = pd.to_datetime(sample['Date '], format="%d/%m/%Y") sample['weeknumber']=sample["Date "].dt.week
results=sample.groupby(['year', 'weeknumber']).agg({'positive':'mean', 'negative':'mean','Text':'size','overall_positive' :'sum','overall_negative':'sum','overall_neutral':'sum'})
results=s.groupby(['year', 'weeknumber']).agg({'positive':'mean', 'negative':'mean','Text':'size','overall_positive':'sum', 'overall_negative':'sum','overall_neutral':'sum'})
Descriptive statistics of the variables.
