Investigating dynamic interdependencies between traditional and digital assets during the COVID-19 outbreak: Implications for G7 and Chinese financial investors

This paper discusses the relationship between the volatilities of traditional and digital assets before and during the COVID-19 pandemic. Using daily data relevant to the period ranging from January 4, 2016, to April 15, 2020, the results of the DCC-MVGARCH model indicate that the stock markets responded to the Coronavirus outbreak as the crypto market with worrying volatility. Before this outbreak, Bitcoin and gold are considered as a hedge for US, English, French, German, and Italian financial investors. The conditional correlation between stock indices and other assets was skyrocketing during this pandemic, except for the couple SSE-Ripple.


Introduction
The template presents the sections that can be used in a manuscript. Each section has an As the COVID-19 virus outbreak news moves far beyond China, the Coronavirus is sending ripples through the stock market and world economy. In just the last few months, this pandemic has managed to plunder and destabilize the world's financial system. For centuries, during numerous crises, gold has traditionally been considered a haven. Since the Bretton Woods crash, gold does not have the same significance in the international monetary system. However, it attracts tremendous interest from investors, the media, and researchers. Due to the growing volatility of financial markets, the diversification of portfolios through hedges is becoming more relevant. During the global economic and financial crisis that started in 2007 with the subprime mortgage market crisis in the United States, gold prices rose dramatically, while other assets suffered losses (Beckmann et al, 2015). In recent years, correlations between most types of assets have increased dramatically. However, gold is still considered to be a zero-beta asset (McCown and Zimmerman, 2006) and is still assumed to be unrelated to other assets (Baur and Lucey, 2010). Bentes (2016) suggested that gold returns had changed from long-range pre-crisis dependency to short-term memory during the crisis period. It also appears that gold is used to act as a hedge in times of financial instability (Selmi et al. 2018;Bouri et al 2020;Jareno et al, 2020;Demir et al, 2020;Snene Manzli, 2021, Kumah et al, 2021;Jeribi and Ghorbel, 2021;Ghorbel and Jeribi, 2021, a,b,c). Dyhberg (2016) asserted that the global uncertainty surrounding the 2007 global financial crisis has eased the emergence of the first decentralized blockchain-based cryptocurrency and strengthened investigated the economic and financial benefits of Bitcoin and Ethereum, for financial investors by studying their capabilities to generate benefits from portfolio diversification and hedging strategies. By estimating static and time-varying tail copulas, they found evidence of time-varying dependence and the absence of asymmetric tail dependence between digital and conventional financial assets. They suggested that digital assets can offer new opportunities for portfolio diversification by including only a small weight of cryptocurrencies in a portfolio of traditional financial assets. Jareno et al. (2020) analyzed the sensitivity of Bitcoin returns to changes in gold price returns, US stock market returns, interest rates, oil prices, and the VIX index. The results indicated that Bitcoin is negatively influenced by VIX, nominal interest rates, and crude oil prices. However, the relations between Bitcoin and gold price returns are positive. Bitcoin is considered a safe-haven asset during economic turmoil. Bouri et al. (2020) compared the safe-haven properties of Bitcoin, gold, and the commodity index against developed, emerging, USA, and Chinese stock market indices. They show that the interdependence between Bitcoin, gold, commodities, and the stock markets is not very strong. They concluded that Bitcoin can be seen as a new virtual gold. It has some of the virtuousness of gold against the extreme drop in the stock market indices.
Bitcoin was created out of the 2008 financial crisis. The Coronavirus pandemic is looking more and more like its first real test in a crisis scenario, as well as the other digital assets. Using the multivariate DCC-GARCH model, we explore the time-varying conditional correlations between stock markets (G7 and China) on the one hand and cryptocurrencies and gold on the other hand. We tested the haven property of cryptocurrencies and Gold from January 4, 2016, to April 15, 2020. The layout of this paper is as follows. Section 2 gives an outline of the econometric methodology adopted. Section 3 is devoted to highlighting the relevant data and empirical findings. Finally, Section 4 concludes.

Empirical Methodology
To determine the nature of the relation between Cryptocurrencies and financial indices of G7 and China, we use the Multivariate GARCH analysis. The use of GARCH models is justified by the presence of heteroscedasticity of the error terms. It is well known that GARCH models offer more dynamic volatility than other models that represent constant variance. Besides, the multivariate GARCH model can present dynamic conditional mean and conditional covariance opposite to the univariate GARCH model. Since the cryptocurrency market is found to be sensitive to related events such as bankruptcy of an exchange or change in regulations, we are using MVGARCH to detect any return volatility within our studied markets. Among all the GARCH models, we are using the DCC-MVGARCH introduced by Engle in 2002. It has been shown in the literature that this model is the most adequate to analyze the cryptocurrency market. It also requires fewer parameters comparing to other GARCH models. Moreover, this upgraded version of GARCH allows more flexibility and is characterized by random walks in returns obtained by pre-filtering and a variance-covariance structure of past returns events. Thus, Structural interdependencies in return-risk are captured by the dynamic behavior of the conditional covariance matrix.
Hence, the first conditional moments (means) of indices returns through a VAR process is presented as follow: Where is a constant, RS,t, and RB,t is the crypto/Gold and stock market conditional mean returns, respectively. These returns are functions of their past market returns, those of the other assets, and their past idiosyncratic shocks. pB and pS are the numbers of autoregressive terms. The choice of VAR order is made based on the parameters significance and Bayesian information criterion (BIC). The residual vector is identically distributed according to a student-t distribution. The structural equation of the MV-GARCH model is as follow: Where σ ² is the conditional variance, α and β represent return volatilities from shocks in own returns reflected respectively by the ARCH and GARCH parameters while ω is a constant. The return volatilities are calculated based on the sum of ARCH and GARCH terms (α +β ) which measures the persistence of volatility during the studied period and must be less than one for systematic volatility to stabilize.
represents coefficients of dummy variables which are included in conditional variance equation to determine volatility estimates relating directly to the COVID-19 pandemic.

Data and Results
The empirical study research involves 1002 daily observations of five cryptocurrencies, Gold, G7, and Chinese market indices, sampled from 4 January 2016 to 15 April 2020. Daily time series data are collected for stock prices of the United States (S&P500), France (CAC40), Germany (DAX30), the United Kingdom (FTSE 100), Italy (MIB 30), Japan (Nikkei 225), Canada (SP-TSX), and China (SSE) from DataStream. From the same database, we extracted the gold prices. Data concerning the five studied cryptocurrencies (Bitcoin, Dash, Ethereum, Ripple, and Monero) was collected from the Coin Market Cap basis. Daily returns are defined by = 100 * ( / ), with , denoting indices i closing price at time t. This transformation of data is usually employed for index series to obtain stationary data. In addition, figures 1 and 2 confirm that the return series are stationary.
Besides, we use two dummy vectors to assign two sub-periods of equal length. D1, t shows crisis dummy preceding the stock market influence of the COVID-19 outbreak taking the value of 1 if t falls in the specific period and 0 otherwise. Likewise, D2, t signifies a crisis dummy for the succeeding period. This specification is helpful since the benchmark period for the comparison of the stock-Cryptocurrency versus Gold association before an onset crisis is not based on the entire sample. Rather, the comparison is based on pre and post-crisis periods of the same time frame, which is expected to reveal the changes in investor's behavior (Baur and Lucey, 2009).
To construct a tractable econometric model, we consider the COVID-19 affected period from January 01 to April 15, 2020, enveloping 73 observations. Further, to examine whether the Crypto versus Bond-stock market relationship for the immediately preceding period was similar or not, we consider the trading days of equal length. Furthermore, to control the relationship, we consider an additional 955 preceding trading days. Note: ***, **, * indicate the estimators' significance at the 1%, 5% and 10% levels, respectively Table 1 presents a summary of statistics of conventional and digital assets' returns. Worth highlighting, the conventional assets returns are lower than those of cryptocurrencies. All assets recorded mean positive returns except the stock market indices of Italy, France, and China. Whereas, for the conventional assets, gold presents the lowest risk. In addition, Bitcoin is the least risky cryptocurrency. More specifically, Ethereum and Monero have the highest average return and standard deviation, which suggests that the higher expected profits are partially balanced by higher volatility. Concerning the normality test, Jarques Bera shows that our variables have very high values which indicates that our data is not normally distributed. This confirms the values of Kurtosis and skewness. Note that negative skewness values are present for all stock indices and Bitcoin. Bitcoin and Monero have the highest excess kurtosis among the cryptocurrencies. The empirical statistics of the Engle (1982) test for conditional heteroscedasticity are significant for all cases suggesting the presence of ARCH effects in returns which justifying our choice of GARCH models.  From Figure 1, we confirm this finding and can observe that the returns of digital assets are more volatile than the returns of the indices (Figure 2), which means that Cryptocurrencies can be defined as highly volatile assets. This result is aligned with Brière et al. (2015), Selmi et al. (2018), Beniki et al. (2019), Symitsi and Chalvatzi (2019), and Baur and Hoang (2020). In addition, Ripple and Monero are among the most volatile; however, Bitcoin is the least volatile of all. We identify high volatility for all cryptocurrencies at the end of 2017, particularly with the launch of Bitcoin futures. This high volatility continued until early 2018, coinciding with the corrections of speculative bubbles that caused an increase in cryptocurrency prices. The crypto market became more volatile after the appearance of the COVID-19 in early 2020. With China and South Korea accounting for more than 70% of Bitcoin's mining capacity, we note that the massive effect was reported as the virus began to spread to Asian countries, much less to the world. It can be shown that digital assets present almost identical movements of volatility. The stock markets responded to the Coronavirus outbreak as a crypto market with worrying volatility, as traders were panic sold out of fear. Indeed, equities have plummeted and stock market volatility has skyrocketed. Moreover, the oil price battle between Russia and Saudi Arabia, and the low price of oil remain the stock markets more volatile. Gold is considered a haven (Baur and Lucey, 2010;Klein et al, 2018;Charfeddine et al, 2020;2021, a, b,c, Jeribi andLahiani et al, 2021;Fakhfekh et al 2021 and investment in times of crisis. Gold is generally less volatile than other commodities, but it was volatile this time. The price of gold continued to rise by mid-February, guided by two main factors. The first was a safe-haven investment resulting from global uncertainty and the second was low to negative returns on highquality government bonds. In mid-March, the price of gold plummeted along with most other commodities and markets. Gold is not playing the role of a haven that it has performed in the past. The DCC model is the preferred model for most of the models tested. However, except for interdependencies between FTSE / CAC40 and Monero / Ripple; Nikkei and Monero/Gold; DAX30 and Ripple; S&P TSX and Dash/Monero. Note that the restricted correlation model (Constant Conditional Correlation) shows less evidence of volatility spillovers. First, the estimated coefficients of DCC(A) and DCC(B) are both positive and statistically significant in all cases. These estimated coefficients sum to a value that is less than one. This implies that the dynamic conditional correlation is mean reverting in all cases. For the CCC model, the correlation coefficients are each positive and statistically significant at the 1% or 5 % level. The highest correlation is between Dash and S&P TSX and the second-highest correlation is between Monero and S&P TSX. The results of the mean equations indicate that the S&P 500 with a lag of one period affects the current value of the cryptocurrency market. However, The S&P 500 with a lag of two periods is not statistically significant for all cases. On the other hand, only the lagged Bitcoin with a lag of one period can affect the current values of the S&P 500. Indeed, the coefficient is statistically significant at the 10% level. Besides, gold with a two lag is statistically significant and affects the S&P 500. For S&P TSX with a lag of one period, the influence is shown only with Bitcoin, Ethereum, and Ripple.
In the next step, the two-dimensions asymmetric DCC-GARCH (1,1) model is estimated, and the estimation results are shown in the second part of Table 2.
are conditional ARCH effects that measure the short-term persistence, while is conditional GARCH effects that measure the long-term persistence. We mainly focus on the GARCH effect caused by heteroscedasticity. For the entire sample, the own conditional ARCH effects are significant at the 1% level, which means that each market is influenced by the short shocks of their own. Meanwhile, the own conditional GARCH effects are also significant at the 1% level. Each market is influenced by the volatility of its market. For each market, the estimated values are smaller than their respective estimated values, indicating that their volatility long-run (GARCH) persistence is larger than short-run (ARCH) persistence.
It is also interesting to find that the effect of the second dummies variable related to the succeeding COVID-19 period is significant. All DCC-GARCH tables confirm that the coefficients of ( ) are positive and significant at a 1% level, for only two cryptocurrencies: Bitcoin and Dash, and for Gold. This result means that the succeeding period of the COVID-19 crisis increases the volatility for the mentioned digital asset and Gold. The coefficients ( ) appear to be significant in all cases except for S&P TSX at least at a level of 10%. Thus, the empirical result shows that the Covid-19 crisis increases the volatility of those stock indices.   Figure 3, we show that the dynamic correlation between Bitcoin and stock market indices (S&P 500, FTSE, DAX 30, CAC 40, FTSE-MIB, and SSE) is negative and low before the COVID-19 pandemic which indicates that the risks among these stock markets can be hedged by Bitcoin. This result is consistent with that of Dyhberg (2016) and Guesmi et al. (2019) and inconsistent with that of Bouri et al. (2017), Kajtazi andMoro (2018), andCherfeddine et al. (2020). Dash and Monero are also considered as hedge assets for US investors before 2020. We suggest also significant benefits for portfolio diversification and risk management when digital assets as Dash, Ethereum, Monero, and Ripple were added. Rounding up the analysis, we explore the relationship between Gold and stock market indices. Based on the results from figure 8 and applying the categorizations of Baur and Lucey (2010), we find that that gold acts as a hedge for US, English, French, German, Italian, and Canadian financial investors. However, it acts as a diversifier asset for Chinese investors. During the COVID-19 outbreak, the conditional correlation between stock indices on the one hand and digital assets and gold, on the other hand, was skyrocketing, except for the couple SSE-Ripple, which confirms the contagion effect of coronavirus between them. Also, Ripple can be considered a haven asset for Chinese investors.

Conclusion
Using data relevant to G7 and Chinese stock market indices, five popular cryptocurrencies (Bitcoin, Dash, Ethereum, Monero, and Ripple), and Gold, we have investigated the relationship between the volatilities of traditional and digital assets. The results of the DCC-MVGARCH model identify an important effect of the COVID-19 outbreak. The crypto market became more volatile after the appearance of the COVID-19 in early 2020. Digital assets present almost identical movements of volatility. The stock markets responded to the Coronavirus outbreak as a crypto market with worrying volatility, as traders were panic-sold out of fear. Indeed, equities have plummeted and stock market volatility has skyrocketed. Gold is generally less volatile than other commodities but it was volatile this time. Gold is not playing the role of a haven that it has performed in the past. When estimating the dynamic conditional correlations, we find, firstly, that the dynamic correlations between cryptocurrencies and stock market indices are low before the appearance of the COVID-19. We also discovered that cryptocurrencies are rather separated from other markets. Dynamic correlations between cryptocurrencies and stock market indices remain high and positive as news of the epidemic COVID-19 virus spreads further beyond China. Before 2020, the results indicate that the risks among the US, English, German, French, Italian, and Chinese stock markets can be hedged by Bitcoin. Dash and Monero are also considered hedge assets for US investors before the Corona virus pandemic. The findings suggest also significant benefits for portfolio diversification when digital assets as Dash, Ethereum, Monero, and Ripple were added. During the first quarter of 2020, the conditional correlation between stock indices and other assets was skyrocketing, except for the couple SSE-Ripple. Also, Ripple can be considered a haven asset for Chinese investors.
The empirical findings have implications for both institutional and retail investors, as well as financial advisers, and explain why risk spillovers in G7 countries and China remain asymmetric throughout mitigated times. As a result, we may deduce two major management implications. First, decision-makers in the G7 and China should support financial markets whenever there is a threat of collapse. Therefore, the risk of a market crash can be better managed during a crisis. Second, particularly during the COVID-19 crisis, investors might use hedging or haven strategies. From an investing point of view, oil prices may impact stock prices in two ways: directly through corporate cash flows and indirectly through the discount rate. This can be considered as a new potential area for further research. Furthermore, cryptocurrencies serve as a means of exchange, a unit of account, and a store of value, all of which contribute to a better understanding of the economy's susceptibility to shocks. As a result, the stock market, cryptocurrency, and commodity behavior are critical to policymaking and boosting economic activity, particularly during a pandemic.
Availability of Data and Materials: Daily time series data are collected for stock prices of the United States (S&P500), France (CAC40), Germany (DAX30), the United Kingdom (FTSE 100), Italy (MIB 30), Japan (Nikkei 225), Canada (SP-TSX) and China (SSE) from DataStream. From the same database, we extracted the gold prices. Data concerning the five studied cryptocurrencies (Bitcoin, Dash, Ethereum, Ripple, and Monero) was collected from the Coin Market Cap basis. In case of request, I provide all the requested data.