Time–frequency analysis of cryptocurrency attention

ABSTRACT

We present a wavelet analysis of retail investor attention and the daily returns of Bitcoin, Ethereum, and Litecoin at five selected crypto exchanges that identifies the fractal dynamics of the short- and long-term persistent processes. The investors’ attention is proxied by the Search Volume Index provided by Google at daily frequency. We detect significant temporal cyclical movements and coherence between cryptocurrency returns and retail investor attention at long investment horizons: from the beginning of 2017 to the middle of 2018 and, to a lesser degree, in 2019. Investment horizons that dominated in 2017 and 2018 were mainly driven by retail investor attention rather than by uncertainty, risk, or stock markets. Therefore, we do not confirm that cryptocurrencies can be considered a safe-haven asset in times of crisis because there is no significant negative comovement between the returns of cryptocurrencies and stock returns or economic uncertainty. Furthermore, the phase shift analysis indicates that attention can serve as a leading indicator for the cryptocurrency returns, particularly in 2017 and 2018. Therefore, retail investors are encouraged to use the Search Volume Index as an early warning indicator in case of sudden changes in the cryptocurrency returns to maximise profits or minimise losses.

KEYWORDS:

• cryptocurrencies
• attention
• Search Volume Index
• wavelet analysis

JEL CLASSIFICATIONS:

• G14
• C14
• C58

Acknowledgements

Svatopluk Kapounek was supported by the Czech Science Foundation via grant No. 20-17044S. We benefited from comments and suggestions by anonymous referees, Frederik Junge, Peter Huber, Evžen Kočenda, Fabian Reck, Florian Horky, Maria Siranova, Puyu Ning, Mihai Mutascu, Camelia Turcu, Makram El-Shagi, and other participants at international scientific conferences.

Table A2. Volumes of selected markets.

Declaration of interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability statement

The data that support the findings of this study are openly available in the Harvard Dataverse, V1 at https://doi.org/10.7910/DVN/N2PMNU.

Notes

1 Source: https://www.netmarketshare.com/search-engine-market-share.aspx?qprid = 4&qpcustomd = 0.

2 The normalised search query index at a given point in time is the ratio of the total search volume for each query to the total number of all search queries. In the robustness analysis, we use the ASVI (Da et al., 2011). Our keywords jointly use the name of a selected cryptocurrency with the name of a cryptocurrency exchange (e.g., ‘Bitcoin Coinbase’) to proxy the demand for the analysed cryptocurrencies.

3 The Search Volume Index is scaled on a range of 0 to 100, based on proportion to all searches on all topics during the specific time period. We employ Google Trends API and download daily data for the individual months. Then, we use monthly data in order to chain daily data using December 1, 2017 as the base period. This base was selected because the number of searches reaches the maximum during the entire analysed period.

4 Due to low liquidity and long transaction periods, daily data is likely to be more appropriate for the presented analysis than high-frequency data.

5 Some smaller crypto markets (e.g., QuadrigaCX and Coinfloor, which are not analysed here) experienced the crash up to two days later in 2018.

6 The Morlet wavelet provides an optimal trade-off between both time and frequency localisation in the financial time series (Crowley, 2007; Rua, 2010). The oscillation is regulated by the parameter ${\omega }_0$, which leads to improved scale localisation and decreased time localisation, and vice-versa. For this analysis, a ${\omega }_0=6$ is chosen, as it exhibits strong similarities to the Fourier period, which improves the interpretation of the result, in accordance with earlier wavelet studies that were conducted in the economic field (Rua, 2010).

7 Note that the smoothing operator $S$ is used again, but in this case, to reduce possible noise in the data.

8 The statistical significance at 5% against white noise is estimated using Monte Carlo simulations.

9 The direction of the leading or lagging time series is represented by arrows (a left arrow denotes the antiphase [180°], while a right arrow denotes the inphase [0° or 360°]). However, the phase as a lead or a lag must be interpreted relative to the antiphase because a lead of 90° is also a lag of 270°.

10 We only present results for one selected crypto exchange (Kraken) in Figures 4–6: these are representative for all analysed crypto exchanges. The additional results are available in Tables A2 and A3 in the Appendix.

Additional information

Notes on contributors

Zuzana Kučerová

Zuzana Kučerová is an associate professor at the VSB-Technical University in Ostrava and Mendel University in Brno, Czech Republic. She completed her PhD degree in economics at the VSB-Technical University in Ostrava. She participates in the research groups that are related to fiscal policy and debt valuation, crowdfunding, interactions between the financial sector and the real economy, and marcoprudential policies, among others. Her research interests include crowdfunding, exchange rates, monetary integration, financial integration, economic policy, and cryptocurrencies. Her work has been published in Economic Modelling, International Review of Economics & Finance, Journal of Behavioral Finance and Baltic Journal of Economics.

Svatopluk Kapounek

Svatopluk Kapounek is an associate professor at Mendel University in Brno, Czech Republic. He completed his PhD degree in finance at Mendel University. He is the leader of the Financial Hub research team at Mendel University and participates in the research groups that are related to fiscal policy and debt valuation, uncertainty in financial markets, and interactions between the financial sector and the real economy, among others. His research interests include international finance, macroeconomics, financial markets, and financial econometrics. His work has been published in Economics Systems, Computational Economics, Economic Modelling, Journal of Behavioral Finance and International Review of Economics and Finance.

Jarko Fidrmuc

Jarko Fidrmuc has gained diverse interdisciplinary experience at several universities and research institutes. He completed his PhD degree in economics at the University of Vienna. In 2005, he was appointed a professor of political economics and European integration at the Economics Faculty and Geschwister Scholl Institute of the University of Munich. Before he joined Zeppelin University as a professor of international economics in 2011, he worked at the Institute for Advanced Studies in Vienna and the Foreign Research Department of the Austrian National Bank. His work has been published in Economic Modelling, Journal of Banking and Finance, Finance Research Letters, Macroeconomic Dynamics, Open Economies Review, European Journal of Political Economy, Regional Studies and Journal of Comparative Economics.