High-dimensional CoVaR network connectedness for measuring conditional financial contagion and risk spillovers from oil markets to the G20 stock system

Highlights

• A high-dimensional CoVaR network based on the LASSO-VAR model is constructed.

• Oil risk spillovers to G20 stocks in crisis period are found from two perspectives.

• G20 stock conditional risk contagion shows regional and oil-related characteristics.

• North American countries are the most affected while Asian countries have few shocks.

Abstract

This paper employs a new framework, the high-dimensional conditional Value-at-Risk (CoVaR) connectedness based on the LASSO-VAR model, to explore the conditional financial contagion among a specific system conditional on extreme events occurring outside the specific system and its extreme risk spillovers to the system from the systemic perspective. Then, this paper employs the delta CoVaR and CoVaR networks to analyse the risk spillovers from oil markets to the G20 stock system from both the pairwise and systemic perspectives. From the pairwise perspective, the delta CoVaR (∆CoVaR) results show that when separating every G20 stock market from the whole market system, there are significant risk spillovers from oil to G20 stocks only during the crisis period. Further, the CoVaR connectedness results show that the G20 stock contagion presents regional characteristics and oil-related characteristics conditional on oil in extreme risk, and also verify the significant risk spillovers from the oil market to the global stock system from the systemic perspective. North American oil-related countries, including the United States of America, Canada and Mexico, are the most affected, and Asian countries have few shocks when the oil market shifts to extreme risk from a normal state. Last, this paper proposes two main policy implications after considering the empirical results.

Introduction

With the development of financial integration, the relationships between financial markets become increasingly closer, and the stability of a financial system decreases dramatically. Modern communication technologies also create favourable conditions for financial risk contagion. The extreme risks of one financial market or institution can spread to other markets or institutions via the open financial market system, causing risk spillovers and even financial contagion. The 2008 global financial crisis and the 2010 European debt crisis both support this phenomenon. The crucial commodity of crude oil especially plays a pivotal role in economic activities and has great relationships with stock markets (Jones and Kaul, 1996; Kilian and Park, 2009; Mensi et al., 2017; Dai et al., 2021; Ma et al., 2021). Thus, this paper models the financial contagion and explores the risk spillovers from the oil market to the stock market system, which can offer some theoretical and practical insights for investors and financial regulators to handle oil shocks, monitor oil risk spillovers, and even prevent systemic risks caused by extreme oil risks.

Regarding the relationships between oil and stocks, previous representative literatures mainly focused on three aspects, i.e., the impacts of oil shocks on stock markets, the co-movement between oil and stock markets, and oil and stock market forecasting. Regarding the impacts of oil shocks on stock markets, the main research methodologies are the vector autoregression (VAR) and the structural vector autoregression (SVAR), and the main conclusions are summarized as follows. First, there are significant negative impacts of oil shocks on the U.S. and most European countries (Sadorsky, 1999; Papapetrou, 2001; Cunado and Perez de Gracia, 2014), mainly because oil is an important productive factor that then affects economic activities. Second, the reactions of the U.S.'s and oil trading countries' stock returns to oil shocks depend on which of supply and demand shocks drive the oil prices (Kilian and Park, 2009; Wang et al., 2013), as well as the corresponding country category, i.e., an oil importer or oil exporter (Wang et al., 2013). As a supplement, Apergis and Miller (2009) point out that stock returns respond little to oil shocks.

Regarding the co-movement between oil and stock markets, the main research methodologies include multivariate generalized autoregressive conditional heteroscedasticity, Copula and extreme value theory, and the main conclusions include the following. First, the correlation or dependence between oil and the stock market for most countries is positive (Aloui et al., 2013; Sukcharoen et al., 2014; Zhu et al., 2014) and around zero in calm periods (Martín-Barragán et al., 2015). Moreover, Conrad et al. (2014) point out that the long-term oil and USA stock correlation is counter cyclical with the sign changes before and during recessions and remains positive throughout the economic recovery. In addition, Zhu et al. (2014) and Sukcharoen et al. (2014) point out that the dependence between oil and stock markets is relatively weak for most cases, but it is relatively strong for large oil consuming and producing countries, e.g., the U.S. and Canada. Second, there are positive extreme dependencies between oil and stock markets for some countries (Aloui et al., 2013; Chen and Lv, 2015; Shahzad et al., 2018). Regarding oil and stock market forecasting, Driesprong et al. (2008) find that a rise in oil prices drastically lowers future stock returns, and Wang et al. (2018) show that oil volatility can predict stock volatility in the short-term.

Over the past decade, the conditional Value-at-Risk (CoVaR) (Girardi and Ergün, 2013; Adrian and Brunnermeier, 2016), systemic expected shortfall (SES) (Acharya et al., 2016), and SRISK (Brownlees and Engle, 2017), as well as financial networks including the Granger causality network (Billio et al., 2012; Basu et al., 2017), the systemic risk beta network (Hautsch et al., 2015), and the variance decompositions network (Diebold and Yılmaz, 2014; Demirer et al., 2018), were proposed for systemic risk analysis, gradually becoming a hot point in the empirical finance area (Castro and Ferrari, 2014; Borri, 2019).

With respect to the concrete topics of oil risk spillovers in energy finance area, the CoVaR and ∆CoVaR empirical results show there are significant asymmetric risk spillovers from oil risks to stock risks for most cases (Mensi et al., 2017; Ji et al., 2020; Li and Wei, 2018; Shahzad et al., 2018). Relative to the CoVaR application in risk spillover or contagion (Mensi et al., 2017; Ji et al., 2020; Wen et al., 2020), the system-wide financial network can describe the interactions of the risks for every financial institution or market, and thus be appropriate to explore the complex systemic characteristics exhibited during a period of extreme market risk (Diebold and Yılmaz, 2014; Hautsch et al., 2015; Zhang, 2017; Demirer et al., 2018; Geng et al., 2021). As far as we know, there are three representative literatures that analyse the interactions among crude oil and a few stocks markets, including Maghyereh et al. (2016), Zhang (2017) and Ma et al. (2019). Maghyereh et al. (2016) construct 12-node network connectedness for oil and stock implied volatility indexes at the country level, finding that the risk transfer between oil and equity markets is asymmetric and dominated by the transmissions from the oil market. Meanwhile, Zhang (2017) constructs 7-node network connectedness for oil and stock returns at the country level, finding that the contribution of oil shocks to the world financial system is limited. Ma et al. (2019) focus on the common and idiosyncratic components of the oil and US energy sector stocks' returns and find that the risk spillovers from oil to US energy sector stocks are mainly driven by common information. Other related papers about oil and stock network connectedness include Antonakakis et al. (2017), Ferrer et al. (2018) and Husain et al. (2019). Antonakakis et al.'s (2017) network connectedness has only one stock, which can hardly describe the interactions among various stock markets; and Ferrer et al. (2018) and Husain et al. (2019) include other financial assets except for oil and stocks.

Particularly, the systemic risk or financial contagion often occurs under extreme market conditions, in which a tail risk can well capture these changes. As far as we know, there are only a few papers that construct tail risk networks to explore financial contagion or extreme risk spillovers (Hautsch et al., 2015; Betz et al., 2016; Härdle et al., 2016; Wang et al., 2017; Corsi et al., 2018). Meanwhile, the contagion triggers always come outside the concerned financial system, and regulators and policymakers often focus on the systemic risks for this system when some extreme event outside the system occurs. For example, oil market is different from global stock markets since oil is a commodity largely affected by geopolitics, indicating the necessity of distinguishing oil and stock markets. In this case, a conditional tail risk network and its changes may be appropriate to explore this conditional financial contagion or risk spillovers inside a financial system conditional on some extreme risks. It is greatly helpful for investors and policy makers to understand the financial contagion laws and then regulate the financial markets to prevent financial systemic risks when extreme events occur outside the specific financial system.

Further, with respect to the variance decomposition network, the number of global main stock markets is more than 20, and the general VAR model cannot meet the requirements due to the curse of dimensionality and estimation error in optimization when there are too many parameters to estimate. Demirer et al. (2018) propose employing the LASSO-VAR model to construct the variance decomposition network to overcome the curse of dimensionality to some degree. Last, the financial contagion characteristics between oil and stocks vary over time, and the financial contagion from oil to stock markets becomes stronger during financial crisis periods, e.g., strong evidence in China and the United States (Wen et al., 2012) and Germany, Japan, the United Kingdom and the United States (Martín-Barragán et al., 2015).

Thus, this paper contributes to the current literature in both methodology and empirical analysis as follows. First, this paper proposes an employing CoVaR network, a conditional tail risk network, to capture the risk spillovers from oil to the global stock system, and then analyses the conditional interactions among all markets of the whole system, i.e., the systemic financial contagion conditional on some specific external situation. Second, this paper introduces the high-dimensional network connectedness based on the LASSO-VAR variance decomposition method, which was proposed by Demirer et al. (2018); and combines it with the Copula-CoVaR method to construct a high-dimensional CoVaR network, which will largely advance the application scope of conditional tail risk networks.

Last, using the G20 stock markets, including G7 developed countries, BRICS developing countries and other important economies, as representative of the global stock market system, this paper divides the time period into three representative macroeconomic periods, i.e., the pre-crisis period, the crisis period and the post-crisis period, for empirical analysis. Particularly, it explores the risk spillovers from the oil market to the global stock system from two perspectives, i.e., the pairwise CoVaR perspective and the system-wide CoVaR network perspective. Both of them conduct the empirical analysis via setting the oil market as conditional variable. It is reasonable and valuable to use the oil market as conditional variable for the two following reasons: (1) The oil market does not belong to the stock system, and its attributes are different from the various stock markets; (2) In our framework, we employ CoVaR to measure the risk spillover from oil to G20 stock market and compare the similarities and differences between G20 VaR network and CoVaR network. By this way, the impact of oil as exogenous shock can be well identified.

This paper has contributed to the existing literature from both methodology and empirical analysis. For the methodology, Demirer et al. (2018) construct the network for stock volatility, which can reflect the risk spillover in some degree. Meanwhile, White et al. (2015) propose VAR for VaR model to measure the tail risk VaR via considering two financial series, and it can also measure the risk spillover for two financial markets. However, combined with stock volatility, VaR contains conditional volatility, conditional mean and tail quantile of standardized distribution for stock series. Thus, VaR connectedness can well measure the tail risk spillovers which considers the extreme market conditions. Meanwhile, CoVaR contains conditional volatility, conditional mean, tail quantile of standardized distribution for stock series, and also contains the joint distribution for stocks and oil. Thus, compared with Demirer et al. (2018)’s volatility connectedness, this paper constructs CoVaR connectedness to reflect the conditional tail risk contagion for stock markets. Particularly, the high-dimensional CoVaR network can also measure the conditional financial contagion and risk spillover among a larger number of financial markets, extending pairwise analysis to a multivariable network framework.

For the empirical analysis, Ji et al. (2020) mainly employed CoVaR based on oil price decomposition to explore the risk spillovers from different types of oil shocks to BRICS stocks. Li and Wei (2018) employed time-varying copula-CoVaR model via combining the variational mode decomposition method to explore the risk spillovers from crude oil to China's stock market, and Wen et al. (2012) similarly employed time-varying copula model to measure the financial contagion between oil and China and United State stock markets. Above literature focused on the pairwise empirical analysis which cannot provide clear evidence for a worldwide market. Zhang (2017) employed network connectedness to explore the risk spillovers among crude oil and six stock markets, while it focused on returns rather than tail risk. Therefore, this paper constructs the high-dimensional VaR network and CoVaR network to analyse the tail risk spillover among the selected G20 stock markets, including G7, BRICS and other important economies.

The rest of the paper is structured as follows. In Section 2, we introduce the methodology. In Section 3, we conduct the empirical analysis. Finally, we conclude in Section 4.