Cryptocurrency price volatility and investor attention

https://doi.org/10.1016/j.iref.2021.06.007Get rights and content

Abstract

This study examines the relationship between the price volatility of cryptocurrencies and investor attention. Using a large dataset of approximately 25 million tweets about 23 of the largest cryptocurrencies, I show that investor attention, as proxied by the number of tweets, retweets, and favorites, corresponds to greater cryptocurrency price volatility. I use a Vector Autoregression (VAR) framework to show that investor attention predicts future price volatility. Additionally, days on which investors are “distracted” because of attention-grabbing events correspond to lower price volatility in cryptocurrency markets. The results suggest that increased investor attention to cryptocurrencies has the undesirable effect of increasing price volatility.

Introduction

Bitcoin, the world's largest cryptocurrency, was the world's best-performing currency (in returns) in 2013, 2015, and 2016. It was also the world's worst-performing currency in 2014 (Desjardins, 2017), only to experience extreme movements again in 2017 and 2018. “Exceptional price volatility” is one of the main undesirable features of blockchains (Saleh, 2019); excessive volatility may be problematic to investors and users of cryptocurrencies. Understanding cryptocurrency price volatility has emerged as an important research topic, especially as cryptocurrencies gain more ground as an investment vehicle and asset class. Particularly, an understanding of cryptocurrency volatility is relevant to portfolio management and risk management. This paper aims to study cryptocurrency price volatility by examining the (causal) impact of investor attention on price volatility; the empirical analyses confirm this relationship. The results can be explained by the theory of investor attention and market volatility developed by Andrei and Hasler (2015) – that volatility increases with attention. The measures of investor attention used in this study are derived from a novel detailed Twitter dataset containing tweets, retweets, and the number of favorites about the largest 23 cryptocurrencies by market capitalization.

Attention is a scarce resource (Kahneman, 1973), and investors suffer from attention deficits when faced with many events competing for their attention. This limited attention, or distraction, causes undesirable market effects, such as greater under-reaction to earnings announcements (Hirshleifer, Lim, & Teoh, 2009). Moreover, investors are slow to react to earnings announcements made on Fridays – presumably when their attention is lowest (DellaVigna & Pollet, 2009). Additionally, Gargano and Rossi (2018) show that attention is positively related to portfolio performance; they suggest that attention is both important and useful. However, other studies indicate that limited attention (or distraction) may positively impact some investors. Peress and Schmidt (2020) find that days on which noise (retail) investors are “distracted” are associated with lower volatility – consistent with such traders mitigating adverse selection, a positive outcome for retail investors.

Whether attention has a positive or a negative impact on cryptocurrency markets is unclear. While crypto markets share some features of the stock market, they also exhibit some important differences. While equity markets have a large presence of institutional investors, cryptocurrencies are unlikely candidates for institutional investors (Białkowski, 2020). The presence of two potentially competing hypotheses of attention, and the unique nature of crypto markets make the question of relating investor attention to price volatility of interest to researchers and investors1 alike; this is especially the case in the already-volatile crypto markets.

In this study, I derive measures of investor attention from a unique large dataset of 25 million tweets about 23 largest cryptocurrencies. The dataset includes information about tweets, retweets, number of favorites, and followers of each tweeting account. This large dataset represents the aggregate opinions of over one million Twitter users worldwide, thus representing a large cross-section of investors who utilize numerous tweeting modes (using iPhone, Android, Twitter website, etc.).

While the first set of tests utilizes this Twitter dataset, the second test exploits plausibly exogenous shocks to investor attention in crypto markets in the form of attention-grabbing events in (the unrelated) equity markets. I focus on days on which many firms release their quarterly earnings reports as a proxy for attention-grabbing events. A large number of earnings announcements taking place concurrently is likely to “distract” investors (Hirshleifer et al., 2009), consistent with investors dividing their limited attention (Kahneman, 1973). The proxy used in this study is similar in spirit to Kempf, Manconi, and Spalt (2017), who use extreme industry returns as a proxy for attention-grabbing events. In this study, I use the number of earnings announcements (specifically days with a very large number of earnings announcements) as a proxy for attention-grabbing events. Additionally, I use extreme price changes in cryptocurrency markets as an additional shock to investor attention. Barber and Odean (2008) argue that extreme price movement is itself an attention-grabbing event.

The main contribution of this study is to show that high investor attention to cryptocurrencies predicts (and causes) greater price volatility. With the introduction of derivatives trading in cryptocurrency markets, an understanding of volatility is in order.2 Two empirical advantages of the analysis are the unique and exhaustive tweeting dataset, and the establishment of causality via plausibly exogenous shocks to investor attention in crypto markets. These results are consistent with the notion that increased attention to cryptocurrencies has the detrimental effect of increasing price volatility. Ibikunle, McGroary, and Rzayev (2020) state that high levels of attention may correspond to an increase in uninformed trading in Bitcoin markets. Those uninformed traders are likely better off when they are “distracted,” as suggested by Peress and Schmidt (2020).

Twitter is a good measure of investor attention because it measures the extent to which investors discuss cryptocurrencies; it can be thought of as a revealed measure of attention. When the number of tweets or discussions about a cryptocurrency is high, it signals elevated attention and vice versa. Shen, Urquhart, and Wang (2019) use Twitter as a proxy for investor attention paid to Bitcoin. Besides tweeting about cryptocurrencies, the number of retweets is another measure of investor attention. By retweeting, an investor spreads the information to more readers (followers) and shows that they have seen and paid attention to the tweet. Similarly, when a user marks a tweet as “Favorite,” they have likely read and paid attention to the content. Thus, two additional measures of investor attention can be derived from Twitter: the number of retweets and favorites. I use these two measures to measure daily investor attention to each cryptocurrency alongside the number of tweets. The emerging literature on social media has identified retweets (Chawla, Da, Xu, & Ye, 2016; Crowley, Huang, & Lu, 2018) and favorites (Crowley et al., 2018), as measures of investor attention.

Discerning investor attention from cryptocurrency price volatility is not simple. This is because attention and volatility may be determined concurrently, and there is no clear direction for causality between the two. It is possible that elevated investor attention can drive up volatility, but greater volatility may also attract greater attention. Furthermore, not all attention is the same; investors can express attention positively or negatively – for example, this can be manifested on Twitter by tweets that are positive or negative in tone.

I deal with the identification problem in two ways. First, I use a vector autoregression (VAR) framework. The VAR system and Granger causality tests allow us to establish predictability (causality) between investor attention and volatility. I construct a Sentiment index, capturing the aggregate daily textual sentiment of all discussions about each currency in the sample set to distinguish between positive and negative attention. I find the role of sentiment to be minimal at predicting price volatility.

Second, I use a plausibly exogenous shock to investor attention in crypto markets to establish causality between investor attention and price volatility. Starting with the premise that investor attention is finite (Kahneman, 1973) and knowing that investors face attention constraints when processing a large number of events in the stock market, such as earnings announcements (Hirshleifer et al., 2009), I suggest that days on which many firms announce their quarterly earnings results will attract the attention of at least a subset of investors – including those investing in cryptocurrency markets. This additional attention paid to equity markets means less attention is paid to crypto markets, given that attention is a finite resource. Importantly, attention paid to equity markets is not expected to affect cryptocurrency price volatility directly; this is consistent with intuition and with the findings of Corbet, Meegan, Larkin, Lucey, and Yarovaya (2018) that cryptocurrencies exhibit relative isolation from other financial and economic assets. Attention in equity markets is expected to affect cryptocurrency price volatility only indirectly through the attention channel.

By studying days with high attention paid to equity markets (thus reduced attention to crypto markets), we can establish causality between investor attention and price volatility of cryptocurrencies. In this study, analyzing the 10% of days with the greatest number of earnings announcements in equity markets shows a reduction in price volatility in crypto markets – thus confirming the causal relationship between attention and price volatility.

This study contributes to the literature focusing on the shortfalls of cryptocurrencies. While blockchains and cryptocurrencies serve an important emerging function in the economy, the literature has identified some important drawbacks. Corbet, Lucey, Urquhart, and Yarovaya (2019) outline several drawbacks of cryptocurrencies, such as pricing bubbles, cyber-criminality, and other illicit use. Foley, Karlsen, and Putnins (2019) show that nearly one-quarter of Bitcoin users are engaged in illegal activities, and as much as 46% of Bitcoin transactions could be part of illegal transactions.3 Benetton, Compiani, & Morse (2021) suggest that the proof-of-work mechanism used to ensure the accuracy of Bitcoin's blockchain, requiring significant energy use, has undesirable effects. Such effects include additional pollution and crowding out other energy uses. Biais, Bisiere, Bouvard, Casamatta, and Menkveld (2018) demonstrate that there are cases in which it is desirable for miners in blockchains to diverge or “fork” to multiple chains or “abandon” parts of the blockchain – thus ultimately undermining the integrity of the blockchain. Moreover, Budish (2018) suggests that there are economic limits to how important Bitcoin can become. Particularly, he states that “Bitcoin would be majority attacked if it became sufficiently economically important e.g., if it became a “store of value” akin to gold.” Budish's argument is sometimes referred to as “Bitcoin's fatal flaw.” While Bitcoin has sometimes been compared to gold, Klein, Pham Thu, and Walther (2018) state that Bitcoin and gold “could barely be more different.”

Biais et al. (2018) develop a model in which the fundamental value of cryptocurrencies is the “the stream of net transactional benefits it will provide.” Their model implies that cryptocurrencies’ returns can be quite volatile. Furthermore, Saleh (2019) asserts that “exceptional price volatility” is one of the main undesirable features of blockchains.

This study is closely related to the literature examining the impact of investor attention on returns and volatility. Investor attention affects the stock market (Da, Engelberg, & Gao, 2011; Vlastakis & Markellos, 2012), foreign exchange markets (Goddard, Kita, & Wang, 2015), and most recently, cryptocurrency markets. In cryptocurrency markets, Urquhart (2018) shows that Google Trends data can be used as a measure of attention to Bitcoin. This attention is driven – in part – by lagged volatility and volume. Moreover, Bleher and Dimpfl (2019) show that while returns are not predictable, Google Trends can predict cryptocurrency volatility. Most recently, Shen et al. (2019) show that tweeting volume predicts realized volatility and trading volume. This study advances this emerging literature by using the largest cryptocurrencies in addition to Bitcoin and by incorporating other measures of revealed investor attention derived from Twitter, such as retweets and favorites, as well as the number of followers of each tweeting account. Moreover, this study exploits shocks to investor attention, in the form of attention-grabbing events, to present more evidence of causality.

The findings of this paper have important implications to investors, portfolio, and risk managers in cryptocurrency markets – markets that are characterized by high volatility. Previous research suggests that cryptocurrency markets are unlikely candidates for institutional investors (Białkowski, 2020), and may thus be dominated by retail investors who are more prone to attention and distraction events (Barber & Odean, 2008). The significant presence of retail investors in crypto markets may be partially responsible for the high volatility observed in these markets. In addition, those retail investors are most likely to be active on non-traditional channels, such as social media. One implication of this is that investors and portfolio managers may find it necessary to turn to emerging and non-traditional media sources where crypto investors may be most active.

This paper proceeds as follows: Section 2 describes the datasets, Section 3 conducts the empirical analyses, Section 4 describes the shock to investor attention, Section 5 conducts additional robustness tests, and Section 6 concludes the paper.

Section snippets

Cryptocurrency data

The cryptocurrency price data are obtained through the Application Programming Interface (API) of the website https://coinmarketcap.com/. The data contain hourly pricing information for each currency in the sample, which are used to calculate the volatility measures. The cryptocurrency price data are supplemented with a large dataset of approximately 25 million tweets about the largest cryptocurrencies by market capitalization between November 16, 2017, and November 5, 2018.

Twitter data collection

The dataset used in

Correlation analysis

I begin the empirical analysis with a simple examination of the correlation coefficients among the variables of interest. The results are reported in Table 3. Notably, the three measures of attention derived from Twitter (number of tweets, retweets, and favorites) are all highly correlated. Moreover, the three measures of Twitter attention are positively correlated with the two volatility measures – with correlation coefficients ranging between 0.06 and 0.16. Finally, the two measures of

Plausibly exogenous shock to investor attention

Thus far, the analysis focused on the relationship between investor attention (as measured by the number of tweets about a coin, retweets, and favorites) and cryptocurrency price volatility. While the VAR framework used for much of the analysis is useful, it is also constructive to examine plausibly exogenous shocks to investor attention to establish causality – beyond what is possible in the VAR framework.

While cryptocurrencies form their own asset class, the literature suggests some links

Number of followers and attention

The central hypothesis of this study is that increased investor attention corresponds to greater price volatility. The empirical tests examined measures of investor attention derived from Twitter, such as the number of tweets, retweets, and the number of times tweets are marked as favorite. These measures indeed correspond to investor attention. However, there is one situation that the analysis thus far does not account for; particularly, not all Twitter users are the same. Some Twitter users

Conclusion

In this study, I document a causal relationship between investor attention paid to cryptocurrencies and price volatility. As cryptocurrency markets are quite volatile, understanding volatility is important to researchers and investors – particularly following the advent of derivatives trading in cryptocurrency markets.

Using a VAR analysis, I show that investor attention predicts future price volatility. I derive measures of investor attention from Twitter – namely, the daily volume of tweeting,

References (56)

  • R. Harris et al.

    Inference for unit roots in dynamic panels where the time dimension is fixed

    Journal of Econometrics

    (1999)
  • G. Ibikunle et al.

    More heat than light: Investor attention and bitcoin price discovery

    International Review of Financial Analysis

    (2020)
  • K. Im et al.

    Testing for unit roots in heterogeneous panels

    Journal of Econometrics

    (2003)
  • Q. Ji et al.

    Realised volatility connectedness among Bitcoin exchange markets

    Finance Research Letters

    (2021)
  • T. Klein et al.

    Bitcoin is not the New Gold – a comparison of volatility, correlation, and portfolio performance

    International Review of Financial Analysis

    (2018)
  • A. Levin et al.

    Unit root tests in panel data: asymptotic and finite-sample properties

    Journal of Econometrics

    (2002)
  • T. Panagiotidis et al.

    The effects of markets, uncertainty and search intensity in bitcoin returns

    International Review of Financial Analysis

    (2019)
  • D. Shen et al.

    Does twitter predict bitcoin?

    Economics Letters

    (2019)
  • A. Urquhart

    What causes the attention of Bitcoin?

    Economics Letters

    (2018)
  • N. Vlastakis et al.

    Information demand and stock market volatility

    Journal of Banking & Finance

    (2012)
  • L. Yin et al.

    Understanding cryptocurrency volatility: The role of oil market shocks

    International Review of Economics & Finance

    (2021)
  • D. Zięba et al.

    Shock transmission in the cryptocurrency market. Is Bitcoin the most influential?

    International Review of Financial Analysis

    (2019)
  • M. Abrigo et al.

    Estimation of panel vector autoregression in Stata

    Stata Journal

    (2016)
  • D. Andrei et al.

    Investor attention and stock market volatility

    Review of Financial Studies

    (2015)
  • W. Antweiler et al.

    Is all that talk just noise? The information content of Internet stock message boards

    The Journal of Finance

    (2004)
  • B. Barber et al.

    All that glitters: The effect of attention and news on the buying behavior of individual and institutional investors

    Review of Financial Studies

    (2008)
  • O. Barndorff-Nielsen et al.

    Econometric analysis of realized volatility and its use in estimating stochastic volatility models

    Journal of the Royal Statistical Society: Series B

    (2002)
  • M. Benetton et al.

    When cryptomining comes to town: high electricity-use spillovers to the local economy

  • Cited by (36)

    • Predictability of crypto returns: The impact of trading behavior

      2023, Journal of Behavioral and Experimental Finance
    • Impact of social metrics in decentralized finance

      2023, Journal of Business Research
    • A K-means clustering model for analyzing the Bitcoin extreme value returns

      2023, Decision Analytics Journal
      Citation Excerpt :

      The study confirmed the presence of price patterns after overreactions i.e., the next day’s price changes were bigger than after normal days in both directions by using a number of parametric and non-parametric tests. Volatility estimation is crucial especially while examining cryptocurrencies [29–31]. Volatility is measured from three different perspectives: historical, implied, and future.

    View all citing articles on Scopus
    View full text