Evaluation of the Symmetrical and Asymmetrical Causality Relationship Between Bitcoin Energy Consumption and Stock Values of Technology Companies

Esta obra está bajo una Licencia Creative Commons Atribución 4.0 Internacional.

Cryptocurrencies have greatly impacted business world with a growing interest in fields ranging from computer science to finance and education (Ghosh et al., 2020:1). From the study of Nakamoto (2008) on electronic payment based on cryptographic system until March 2, 2022, 17,938 cryptocurrencies have emerged (CoinMarketCap, 2022). The explosion in the number of cryptocurrencies has offered speculators and investors a diverse range of electronic crypto assets to trade in (Wei, 2018:21). Despite this, Bitcoin remains the leader in the field of crypto assets in terms of market capitalization, user base and popularity (Mikhailov, 2020:89). Bitcoin rules are designed by engineers, it is a digital currency that is not affiliated with any central bank or public enterprise (Böhme et al., 2015:213).

The Bitcoin system is based on a network of users communicating with each other using the Bitcoin protocol via the internet. The Bitcoin protocol is an open source software application and allows users to store and transfer Bitcoins for purchasing and selling goods, or to exchange Bitcoins for other currencies (Vranken, 2017:1). The process of producing Bitcoin, called Bitcoin mining, uses Blockchain technology and basically only requires hardware and electricity consumption (Gallersdörfer et al., 2020:1843). The increase in Bitcoin prices has also led to an increase in the demand for Bitcoin mining. However, both the business world and researchers started to discuss the energy consumption in Bitcoin mining (Küfeoğlu and Özkuran, 2019:1).

Due to the increasing popularity of Bitcoin mining, the operation of the Bitcoin net- work as a whole with more hardware requires much more energy consumption (De Vries, 2021:509). It was concluded that as of June 2018, Bitcoin mining is no longer profitable for electricity prices above 0.14 $/kWh. For this reason, many Western miners have given up on Bitcoin production and the centralization of mining activity in China has increased even more (Delgado-Mohatar et al., 2019:1). The global Bitcoin mining map for August 2021 is shown in Figure 1. In the map, it is seen that the United States (35.40%), Kazakhstan (18.10%), Russian Federation (11.23%) and Canada (9.55%) are the countries with the most Bitcoin mining. After Bitcoin mining was banned in China in June 2021, some of the miners sold their hardware, while others took refuge in places like Texas or Kazakhstan. For this reason, while Bitcoin mining in China is not seen in Figure 1, Bitcoin mining in the United States and Kazakhstan has increased (Shen and Galbraith, 2021).

Energy consumption is an important input for Bitcoin mining based on blockchain technology (Hayes, 2017:1316). Therefore, the increased energy consumption for Bit- coin mining has started a new discussion about the sustainability of the cryptocurrency. On the other hand, most studies ignore the growing amount of short-lived hardware that is one of the important inputs for Bitcoin mining (De Vries and Stoll, 2021:1). In May 2021, approximately 2.9 million specialized hardware devices worldwide were used for Bitcoin mining, generating 160 quintillion guesses per second and consuming approximately 13 gigawatts (GW) of electricity (De Vries et al., 2022:1). Therefore,

Figure 1.
Bitcoin Mining Map
Source: Cambridge Centre for Alternative Finance, 2022.

the increasing electricity consumption for Bitcoin mining and the need for short-lived hardware increase the demand for technology companies. This situation enables the global technology companies, which are the subject of the study, to increase their market shares. Therefore, increasing demand positively affects the stock values of technology companies. In this study, the causality relationship is examined between Bitcoin energy consumption and stock values of technology companies using the Toda and Yamamoto causality test and Hatemi-J asymmetric causality test. With this aspect, the study will contribute to this gap in the literature.

The rest of this study is created as follows: In the subsequent section a brief literature review is given. In section 3, detailed explanations are given about the Toda and Yamamoto causality test and Hatemi-J asymmetric causality test approaches. In section 4, after describing the data, the causality analyses are made and the empirical findings are discussed. Finally, the results are evaluated and suggestions for next studies are presented.

When the literature is searched, there are many studies showing that there is a causality relationship between Bitcoin and other indicators. However, due to the increasing demand for electricity in Bitcoin mining, it has been determined that studies examining the causality relationship between Bitcoin electricity consumption and other indicators intensified in the period after 2018. For this reason, the studies published on the subject after 2018 are summarized in Table 1. In general, it has been determined that there is a causality relationship between Bitcoin electricity consumption and other variables used in the studies.

Table 1.
Literature Summary

In this section, information is given about the Toda and Yamamoto causality test and Hatemi-J asymmetric causality test used in the study. The methodological framework of the study is shown in Figure 2.

Figure 2.
Methodological Framework

3.1.Toda and Yamamoto Causality test

VAR (Vector Autoregressive Models) analysis in time series is a type of analysis that researchers use very often. However, hypothesis tests are not valid in cases where the variables analyzed with VAR are not stationary (if they contain a unit root). After per- forming VAR analysis with series that are stationary (without unit root), According to the Granger causality test, the F statistic is used. However, Toda and Yamamoto (1995) state that if there is cointegration between the variables, the F statistic may not comply with the standard distribution and it may lose its validity. They also stated that VAR analysis can be done by using the level values of the variables and Wald test can be used if the variables contain unit root. In this case, if there is cointegration between the variables, there will be an error correction system (ECM). However, the degree of cointegration of the variables or whether they are cointegrated and whether they have unit roots is not known beforehand. For this reason, many pre-tests are required in the Granger Causality Test. In the Toda and Yamamoto (1995) test, on the other hand, a causality relationship can be tested when the variables become stationary at different levels.

In the Toda and Yamamoto (1995) test, the appropriate lag length (k) is determined using the VAR model. Following this step, the degree of integration (dmax) for the variable with the highest integration degrees and the lag length (k) calculated for the model are added. Finally, the estimation of the VAR model is made by considering the level values (k +dmax) of the series and the delays. The VAR model is implemented through equation (1), (2) (Toda and Yamamoto, 1995:230).

The hypotheses in the model are as follows:

H₀: The X variable is not the Granger cause of the Y variable.

H₁: The X variable is the Granger cause of the Y variable.

The success of the model is associated with the correct creation of the value of the (dmax) and (k) of series.

3.2.Hatemi-J Asymmetric Causality Test

In asymmetric causality analysis tests, it is argued that there is actually a hidden relation- ship between dual-time series, which cannot be correlated at first glance, and that there is no relationship between them, and these hidden relationships can only be found by considering the asymmetry between the components. The Hatemi-J (2012) test allows to investigate this relationship.

In the case of the run the causality relationship between two integrated variables y₁t and y₂t is as follows (Hatemi-J, 2012:449-450);

In the equations, t = 1, 2, …, T, indicates the constant terms, y_1t and y_2tshow funda- mental values, ε_1iand ε_2i are error term. Positive-negative shock is shown in equation (4);

Therefore, it is expressed as

The rewritten form of the equations is as follows;

Lastly, the positive and negative shocks in each variable are expressed in cumulative form as:

Then, assuming that the causality relationship between the positive components are analysed (VAR). VAR (p) model is termed as in equ (8)

Here, reveals a variable vector of size 2x1 v is constant variable vector of size 2x1, is error term of size 2x1.

The symbol in equation 9 represents the j lag length of the calculated VAR model error terms

First, the lag length is determined and then the (W) statistic is used to measure 11) the null hypothesis basic hypothesis, which shows that there is no Granger causality between the series. On the other hand, the VAR models established to reach (W) statistical values are as follows:

with a clearer expression Y = DZ + δ

(10) equation the matrices are of different size Y:(n x T), D: (n x (1 + np)), Zt: ((1 + np) x 1), Z: ((1 + np) x T) and δ: (n x T). these indicators point to different sized matrices

In this study, the causality relationship is examined between Bitcoin energy consumption and the stock values of technology companies. Apple, Dell Technologies, Lenovo Group, HP, Quanta Computer, Compal Electronics, Canon, Wistron and Hewlett Packard Enterprise have been taken into account to represent technology companies. In the analysis, weekly price data for the period 12.02.2017 -07.02.2021 were used. Bitcoin energy consumption data were obtained from the ‘digiconomist.net’ address and are expressed in TWh. The stock value data of technology companies were taken from ‘investing.com’ and are in US dollars. In the analysis, the natural logarithmic values of the data were used. In order to determine the relation between Bitcoin energy consumption and technology companies’ stock value, Carrion-I Silvestre et al. (2009) unit root test with multiple structural break, Toda and Yamamoto (1995) symmetric causality test and Hatemi-J (2012) asymmetric causality test were used.

The time path graphs of the variables are given in Figure 3 for a preliminary assessment of the variables.

According to Figure 3, it is seen that except Canon and Wistron companies show a decreasing trend and there are many structural breaks in each series.

According to descriptive statistic information in Table 2, it is seen that Rquant has the highest average, Rlenovo has the lowest average; Benergy has the highest volatility, and Rcompal has the lowest volatility. When the skewness coefficients are evaluated, it is determined that the distribution of Benergy, Rhp, Rquant and Rhewltent are skewed to the left, and according to the kurtosis coefficients, except Benergy and Rlenovo, series are flattened.

In order to determine the stationarity of the series in the first stage of the research, Carrion-i Silvestre et al. (2009) multiple structural break unit root test was applied. The results of the test are given I Table 3.

Figure 3.
Time Path Plots of Variables

Table 2.
Descriptive Statistics

Table 3.
Carrion-i Silvestre et al. (2009) Unit Root Test Results with Multiple Structural Breaks

Note: PT, MPT, MZα, MSB and MZT are the test statistics of Carrion-i Silvestre et al. (2009) unit root test.

According to Carrion-i Silvestre et al. (2009) unit root test results with multiple structural breaks in Table 3, it is seen that the level values of all series are smaller than the critical values of the test statistics, so that there is no unit root in all series, that is, the series are stationary in their level values.

To determine whether there is a symmetric causality relation between the series, the Toda and Yamamoto (1995) causality test was performed. The results of the test are given in Table 4.

Table 4.
Toda and Yamamoto (1995) Causality Test Results

Note: *** in the table denotes 1% significance level.

According to the results of the Toda and Yamamoto (1995) causality test in Table 4, a causality running from Bitcoin energy consumption to Apple’s stock value appears to be. Thus, Bitcoin energy consumption has been determined to be the Granger reason of the Apple company.

Hatemi-J (2012) asymmetric causality test was used to determine whether there is an asymmetric causality relation between the series. The results of testing are given in Table 5.

According to the Hatemi-J (2012) asymmetric causality test results in Table 5, it has been determined that there is a causality from Bitcoin energy consumption positive shocks to Apple, Dell Technologies, Lenovo Group, HP, Quanta Computer, Compal Elec- tronics, Canon, Wistron and Hewlett Packard Enterprise stock values negative shocks. In addition, it is found that there is a causality from Bitcoin energy expenditure negative shocks to Hewlett Packard Enterprise negative shocks. With these results, Bitcoin energy consumption is the Granger cause of Apple, Dell Technologies, Lenovo Group, HP, Quanta Computer, Compal Electronics, Canon, Wistron and Hewlett Packard Enterprise companies’ stock values.

Table 5.
Hatemi-J (2012) Asimetric Causality Test Results

Note: ***, ** and * in the table denotes respectively, 1%, 5% and %10 significance levels

In recent years, the cryptocurrency markets and the stock exchanges established depending on these markets, the functioning of these stock exchanges, the regulations, the relations of cryptocurrencies with other financial assets and sectors have been the subject of discussion. Especially after 2018, Bitcoin energy consumption based on crypto mining has become the focus of these discussions. For Bitcoin mining, first of all, energy consumption and computer hardware are needed. The demand for computer hardware required for Bitcoin mining plays an increasing role on the stock returns of technology companies. Accordingly, the existence of a relationship between the energy spent for Bitcoin and computer hardware has been wondered. This study focuses on the causal relationship between Bitcoin energy consumption and the stock value of technology companies. The companies (Apple, Dell Technologies, Lenovo Group, HP, Quanta Computer, Compal Electronics, Canon, Wistron and Hewlett Packard Enterprise) that make up the data set of the research are in the computer and office equipment sub-sector of the technology sector in the 2021 Fortune Global 500 list.

In this study, which examines the causal relationship between Bitcoin energy consumption and Apple, Dell Technologies, Lenovo Group, HP, Quanta Computer, Compal Electronics, Canon, Wistron and Hewlett Packard Enterprise, it is taken into account to represent the stock markets of technology companies. Daily price data for the period 12.02.2017 -07.02.2021 were used in the analysis. Toda and Yamamoto (1995) symmetric causality test and Hatemi-J (2012) asymmetric causality test were used to determine the relationship between Bitcoin energy consumption and the stock values of technology companies. According to the results of the Toda and Yamamoto (1995) causality test, it has been found that there is a causality from Bitcoin energy consumption to Apple’s stock value; according to the Hatemi-J (2012) asymmetric causality test results, it has been determined that there is a causality from Bitcoin energy consumption positive shocks to Apple, Dell Technologies, Lenovo Group, HP, Quanta Computer, Compal Electronics, Canon, Wistron and Hewlett Packard Enterprise stock values negative shocks and from Bitcoin energy expenditure negative shocks to Hewlett Packard Enterprise negative shocks. In terms of future studies, it is suggested to create an equation that takes into account Bitcoin energy consumption and state sanctions, the perspective and popularity of global markets on these transactions. The creation of this broad equation will allow to look at the subject from a broader perspective. Finally, it is thought that these findings will provide a market information for international investors, and it has been seen that the profitability of this sector is also closely related to Bitcoin energy consumption.

According to the findings obtained from the study, it is seen that the change in Bitcoin energy consumption has an effect on the returns of the companies that sell the necessary tools for Bitcoin energy production. Because of these results, it was concluded that Bitcoin energy consumption is not only a modern challenge in the new world order, but also a dynamism in the Bitcoin mining industry. In the academic literature research, no study has been found that examines the relationship between Bitcoin energy consumption and the share values of technology companies. In this respect, it is thought that the study is original and will contribute to the literature. Since in the cryptocurrency markets there is not only the energy consumption of Bitcoin but also the energy spent for Ethereum, the causality relationship between the energy consumption spent for Ethereum production and the share values of technology companies can be investigated in future studies. It is also suggested to create an equation that takes into account Bitcoin energy consumption and government sanctions, the perspective and popularity of global markets on these transactions. The creation of this broad equation will allow to look at the subject from a broader perspective.