Dynamic volatility spillovers during market turbulence: examining the interconnectedness among gold, oil, cryptocurrencies and stock markets

Periods of heightened market turbulence, such as the COVID-19 pandemic and the Russia–Ukraine conflict, have renewed interest in how financial shocks propagate across asset classes and markets. During such times, investors seek assets that can shield their portfolios from sharp equity market declines. Hedging instruments are particularly important in these contexts as they help preserve value and reduce volatility when market conditions deteriorate. These assets are crucial for diversification and capital protection, especially when traditional investment strategies become less effective due to rising correlations across markets.

Traditionally, gold and oil have played this defensive role. Gold is widely regarded as a store of value and a hedge against inflation and geopolitical uncertainty. Its historical performance, during the global financial crisis (Baur and Lucey, 2010; Baur and McDermott, 2010), the COVID-19 pandemic (Akhtaruzzaman et al., 2021; Salisu et al., 2021; Hung and Vo, 2021), and other periods of distress, has reinforced its reputation as a safe haven. While sometimes seen as a hedge against inflation and a proxy for global demand, oil has shown more complex behavior. In recent years, its increased sensitivity to macroeconomic shocks and financial market integration has made its hedging effectiveness more inconsistent (El-Chaarani, 2019; Perifanis and Dagoumas, 2021).

The rise of digital finance has introduced new potential hedging assets, most notably Bitcoin. Promoted as “digital gold”, Bitcoin is decentralized, scarce by design and structurally independent from traditional financial systems. Advocates point to its low correlation with traditional assets as a sign of its diversification potential (Bouri et al., 2017). However, its extreme volatility, speculative dynamics and regulatory uncertainty have raised questions about its reliability in times of stress (Wang et al., 2022). While Bitcoin often appears loosely correlated with other markets in stable periods, its behavior during crises is far less predictable. Shahzad et al. (2020) note that Bitcoin's hedging performance depends on the market environment and investment horizon. Recent studies suggest that asset correlations tend to tighten in times of heightened volatility, weakening the hedging value of both traditional and alternative assets (Ghorbel and Jeribi, 2021; Ghorbel et al., 2022; Kayral et al., 2023).

These evolving dynamics raise several important questions regarding the effectiveness of gold, oil and Bitcoin as hedges or safe havens during turbulent periods, the behavior of these assets across markets with varying levels of financial development, and the extent to which volatility shocks in one market transmit to others during times of crisis. Addressing these questions requires a closer examination of volatility spillovers across assets and markets, as they provide valuable insight into how financial shocks are transmitted and whether so-called safe-haven assets retain their protective roles during periods of systemic stress.

This paper investigates the dynamic volatility spillovers among gold, oil, Bitcoin and stock markets in the USA, Japan, China and Vietnam. These assets and markets were selected to capture both traditional and alternative hedging instruments across economies with varying levels of financial development and integration. Gold and oil represent conventional safe-haven assets, while Bitcoin reflects a new and increasingly popular digital alternative. The equity markets span from advanced economies (the USA and Japan) to large emerging (China) and frontier markets (Vietnam), allowing us to evaluate how volatility is transmitted across different levels of market maturity and financial integration. The US market, represented by the S&P 500, is one of the most liquid and globally influential. Japan's Nikkei 225 reflects a developed Asian market with strong links to global capital flows. China's Shanghai Composite captures the dynamics of a large but policy-sensitive emerging market, and Vietnam's VN Index represents a frontier market that has recently seen rapid growth in retail investor participation but remains relatively segmented from global finance. The analysis covers the period from January 2018 to December 2023, encompassing several major episodes of global turbulence, including the COVID-19 pandemic and the Russia–Ukraine conflict.

The spillover index framework developed by Diebold and Yilmaz (2012) is employed to quantify the extent and direction of volatility transmission. This methodology is well-suited for capturing time-varying connectedness and identifying the dominant sources and recipients of volatility across asset classes and regions. Our approach allows us to assess whether traditional and digital assets maintain their hedging roles during periods of stress and how financial shocks propagate differently in developed versus emerging markets.

This paper makes three significant contributions to the literature. First, we offer a unified empirical framework incorporating both traditional and alternative assets. This allows us to evaluate whether their roles as hedging or safe-haven instruments are held across different market conditions. Second, we compare volatility spillovers across equity markets at varying stages of financial development, from highly integrated developed markets like the USA and Japan to more segmented emerging and frontier markets like China and Vietnam. This comparative angle helps clarify how market structure influences the dynamics of volatility transmission. Third, by focusing on recent global events, including the COVID-19 pandemic and the Russia–Ukraine conflict, our study provides timely evidence on whether the presumed diversification benefits of gold, oil and Bitcoin remain valid during periods of systemic stress.

Three key findings are reported. First, volatility spillovers across asset classes tend to surge during major global shocks such as the COVID-19 pandemic and the Russia–Ukraine conflict. This underscores the rising interconnectedness among commodities, cryptocurrencies and stock markets in times of crisis, weakening the traditional diversification benefits. Second, developed markets, especially the USA, emerge as primary sources of volatility transmission, whereas emerging markets, including China and Vietnam, play a more limited role in propagating shocks. Third, while commodities and cryptocurrencies generally exhibit low levels of spillovers to and from other markets, supporting their function as risk diversifiers, their role diminishes during periods of extreme stress. During crisis episodes, these assets become net transmitters of volatility, casting doubt on their reliability as safe-haven instruments when they are most needed.

The rest of the paper is structured as follows. Section 2 reviews the literature on volatility spillovers and safe haven assets. Section 3 outlines the methodology. Section 4 describes the data. Section 5 presents the empirical findings. Finally, Section 6 concludes the study with a discussion of portfolio management and market regulation implications.

The modern portfolio theory, introduced by Markowitz (1952), laid the groundwork for understanding how investors can optimize their portfolios by balancing risk and return through diversification. The core idea is that holding a mix of imperfectly correlated assets can reduce overall portfolio risk without sacrificing expected returns. Diversification is most effective when asset returns respond differently to economic shocks, allowing gains in one asset to offset losses in another. As such, investors increasingly look beyond domestic equity markets, seeking exposure to international markets and alternative assets that offer higher returns or risk-reducing properties.

Emerging and frontier markets have attracted investor interest because of their potential to deliver higher returns and provide diversification benefits stemming from their different economic and financial cycles. According to Bekaert and Harvey (1997), these markets offer diversification gains because of their lower correlations with developed markets, especially during normal periods. However, such benefits tend to diminish during periods of global financial distress, when correlations across markets rise and volatility spillovers intensify (Forbes and Rigobon, 2002). Understanding how volatility and shocks are transmitted across markets becomes critical to reassessing diversification strategies under such conditions.

To manage risk during turbulent periods, investors turn to assets perceived as safe havens or effective hedges. Baur and Lucey (2010) distinguish between hedges (assets that protect portfolios in general) and safe havens (assets that protect during market stress). Gold has long served this role with its intrinsic value and independence from monetary policy. Empirical findings support gold's safe-haven function across multiple crises, including the global financial crisis (Baur and McDermott, 2010), the COVID-19 pandemic (Akhtaruzzaman et al., 2021; Salisu et al., 2021) and geopolitical events such as the Russia–Ukraine war (Havidz et al., 2023). However, gold's safe-haven status is not universal. Evidence from emerging markets is mixed, with some studies suggesting that gold acts more as a diversifier than a consistent hedge (Aftab et al., 2018; Wen and Cheng, 2018).

Crude oil, another major commodity, is often regarded as a macroeconomic hedge due to its link to global economic activity and inflation. However, its effectiveness as a hedge is context-dependent. Oil prices are highly sensitive to supply disruptions, geopolitical tensions and broader market sentiment, which can amplify rather than mitigate risk in certain conditions (El-Chaarani, 2019; Perifanis and Dagoumas, 2021). Thus, while oil can contribute to portfolio diversification, its hedging role is inconsistent across time or market regimes.

The rise of digital assets, especially Bitcoin, has introduced a new dimension to the safe haven debate. Early studies (Bouri et al., 2017; Dyhrberg, 2016) found weak correlations between Bitcoin and traditional financial assets, supporting its role as a diversifier. However, its high volatility, speculative behavior and regulatory risks raise doubts about its stability during crises. More recent evidence suggests that Bitcoin may function more as a speculative asset or short-term diversifier rather than a stable hedge (Ghorbel and Jeribi, 2021; Ghorbel et al., 2022; Kumar and Padakandla, 2022; Kayral et al., 2023).

To evaluate these dynamics more systematically, researchers have increasingly turned to methodologies that measure volatility spillovers and connectedness across markets and assets. Early studies relied on conditional correlation derived from GARCH processes and cointegration approaches, but these are limited in capturing directionality (Masih and Masih, 1999; Miyakoshi, 2003; Bekaert et al., 2005). More advanced frameworks include copula models, wavelet decompositions and quantile-based techniques (Kumar and Padakandla, 2022). Among the most influential methods is the spillover index introduced by Diebold and Yılmaz (2009, 2012). Based on forecast error variance decomposition from a VAR model, this approach quantifies total, directional and net volatility spillovers across assets or markets. Its appeal lies in its ability to capture how shocks are transmitted and received, providing insights into the evolving role of various assets. Numerous studies have adopted and extended this approach to study commodities (Golitsis et al., 2022), cryptocurrencies (Hung, 2022) and stock markets (Wang et al., 2022), often revealing that asset behavior shifts dramatically during crisis periods. Extensions of the Diebold–Yılmaz framework include time-varying parameter models (TVP-VAR), rolling windows and spectral decompositions (Baruník and Křehlík, 2018), allowing researchers to detect short- and long-term spillover patterns and structural breaks during major global events. These methodological advances have significantly deepened our understanding of cross-asset volatility dynamics. They reveal that connectedness is not static but evolves with global events, market sentiment and investor behavior. This evolving spillover structure has important implications for portfolio diversification, systemic risk monitoring and the assessment of alternative assets' hedging effectiveness under different economic regimes.

Despite a growing literature on volatility spillovers and hedging assets, several key gaps remain. First, most studies examine the behavior of gold, oil and Bitcoin in isolation or asset pairs, with limited attention to how these assets interact collectively within broader market systems. Second, few studies use a unified analytical framework to compare their roles across both developed and emerging markets. The safe-haven status of these assets is often assumed rather than empirically tested across multiple crises and market structures. Third, within the same dataset, existing research rarely spans multiple recent global stress events, such as the COVID-19 pandemic and the Russia–Ukraine conflict. This study addresses these gaps by applying the Diebold-Yılmaz (2012) spillover framework to examine dynamic volatility connectedness among gold, oil, Bitcoin and stock markets in the USA, Japan, China and Vietnam from 2018 to 2023. This six-year period captures several major shocks and offers a rich environment to reassess whether traditional and digital assets continue to serve as effective hedging instruments under stress. By including stock markets across different levels of development, this study also provides comparative insights into how financial structure and market maturity influence volatility transmission. Our findings contribute to a deeper understanding of how the roles of gold, oil and Bitcoin evolve in response to systemic risk and whether their diversification benefits persist when investors need them most.

In this study, we analyze cross-market transmission mechanisms by applying the spillover methodology originally proposed by Diebold and Yilmaz (2012). Their framework enables us to track how disturbances arising in one asset class or market propagate through the broader system over time. The approach relies on breaking down the contribution of each shock to the h-step-ahead forecast-error variance within a system of N variables described by an n-dimensional Vector Autoregressive (VAR) model. Let the n-dimensional time series variables $yt=(y1t,y2t,…,,ynt)′$ be represented as a vector process that follows a VAR(⁠$p$⁠) process of the form:

where the coefficient matrices $Φi$ are $n×n$ parameters and $εt$ denotes an independently and identically distributed innovation term.

Under the condition of covariance stationarity, this VAR can be equivalently expressed as its infinite-order moving-average (MA) representation:

where the matrices $Ai$ are obtained recursively from $At=Φ1Ai−1+Φ2Ai−2+…+ΦpAi−p$ and $A0=In$⁠.

The MA coefficients, and the objects derived from them, such as impulse-response functions or forecast-error variance decompositions, capture the dynamic interactions among the variables and form the basis for the spillover analysis employed in this study. Through variance-decomposition techniques, we can identify how much of the future uncertainty in each variable stems from its own shocks as opposed to disturbances originating from other markets.

Following the Generalized VAR (GVAR) methodology developed by Pesaran and Shin (1998) and Koop et al. (1996), the $H$-step-ahead generalized forecast-error variance decomposition matrix $DH={dijH}$ is defined such that each element represents the contribution of innovations in variable j to the $H$-step-ahead forecast variance of variable $i$⁠:

where $Σ$ denotes the covariance matrix of the innovation vector in the non-orthogonal VAR, $Ah$ is the $h$-step MA coefficient matrix, and $ei$ is a unit-selection vector with a one in the $i$-th position and zeros elsewhere. The term $σjj$ represents the square root of the $j$-th diagonal entry of $Σ$⁠, ensuring appropriate scaling of shocks. A key advantage of the generalized VAR (GVAR) framework is that it evaluates spillovers without imposing orthogonality on the shocks. This feature allows the transmission of innovations to be assessed in a way that is fully consistent with the observed error-covariance structure, yielding a more robust and data-driven characterization of cross-market linkages than approaches relying on orthogonalized impulse responses.

Because the shocks in the generalized framework are correlated rather than orthogonal, the rows of $DH$ do not necessarily sum to one. Therefore, each element of the variance decomposition matrix is rescaled by its corresponding row total to ensure proper normalization $d~ijH=dijH∑j=1NdijH.$ By construction, $∑j=1Nd~ijH=1$ and $∑i,j=1Nd~ijH=N.$

The forecast error variances of each variable in the VAR system are divided into different parts based on the variance decomposition. The diagonal elements of $D~H$ explain the fractions of the H-step-ahead error variances in forecasting $yi$ caused by its shocks, known as their own variance shares. On the other hand, the off-diagonal elements represent cross variance shares or spillovers, which indicate the fractions of the H-step-ahead error variances in forecasting $yi$ that are influenced by shocks from other markets.

Diebold and Yilmaz (2012) utilize the GVAR framework to assess the spillover from one market to others (called directional spillover), thereby presenting a significant advancement from their previous work in Diebold and Yilmaz (2009). The directional spillover from market j to market i is represented as $Si←jH=d~ijH$⁠. In general, $Si←jH≠Sj←iH$ because a disturbance originating in market $j$ and influencing market $i$ may not exert the same magnitude or direction of effect as a shock moving in the opposite direction. Consequently, there exist $N(N−1)$ ordered pairs of directional spillovers. The net bilateral spillover from $j$ to $i$ can therefore be expressed as the difference $Si←jH−Sj←iH$⁠.

The total directional spillovers from j to other markets and from other markets to i are presented as:

For each market, there are $N$ directional measures capturing spillovers received from all other markets (“from others”) and another $N$ measures reflecting spillovers transmitted to other markets (“to others”). The net total spillover position of market $i$⁠, denoted $SiH$⁠, is determined by the difference between its outgoing and incoming spillovers: $SiH=S·←iH−Si←·H$⁠.

To capture the overall interconnectedness within the system, Diebold and Yilmaz define a total spillover index by summing all off-diagonal elements of the normalized variance decomposition matrix $D~H$⁠. This metric aggregates cross-market influences and expresses them as a percentage of the total forecast-error variance:

A key advantage of the Diebold–Yılmaz spillover index is its ability to incorporate a rolling window estimation, allowing researchers to capture the evolution of shock transmission over time and across different episodes of crises. This creates a comprehensive and interpretable framework for analyzing both total system-wide and pairwise spillovers dynamically. In contrast, many alternative methods, such as principal component analysis (PCA) or CoVaR, either lack a temporal dimension or focus narrowly on tail risks of individual entities, offering a more limited view of systemic interconnectedness.

To analyze the volatility spillovers across asset classes and stock markets, we use daily data from 01 January 2018 to 31 December 2023. This six-year window captures multiple market behavior phases, including both relatively calm periods and episodes of elevated uncertainty, such as the COVID-19 pandemic and the Russia–Ukraine conflict. These events offer a rich empirical setting to examine how asset connectedness evolves over time and under different stress regimes.

Our sample includes four stock market indices representing economies at different stages of development: the S&P 500 (USA), the Nikkei 225 (Japan), the Shanghai Composite Index (China) and the VN Index (Vietnam). These indices allow us to compare volatility spillovers between developed, emerging, and frontier markets. In addition to equities, we include three widely followed alternative assets: gold (Gold Futures), crude oil (Crude Oil (WTI) Futures) and Bitcoin. To ensure comparability, all series are synchronized to account for differences in trading days and time zones.

We concentrate our analysis on volatility spillovers. Given that volatility is widely interpreted as a proxy for market fear and uncertainty, volatility connectedness can accordingly be viewed as a measure of fear transmission across markets (Demirer et al., 2018). As volatility is latent and not directly observable, we estimate it using the daily range-based measure proposed by Parkinson (1980). The daily variance $σ~it2$ of asset i on day t can be estimated as follows:

where $pitmax$ and $pitmin$ are the maximum and minimum prices of asset i on day t, respectively.

The descriptive statistics for the dataset are given in Table 1. In general, asset volatilities are non-normal, leptokurtic and, in most cases, positively skewed. Bitcoin demonstrates the highest level of excess volatility with an average volatility of 62.26%, followed by crude oil and gold. While still highly volatile, stock market indices show a more moderate range of volatility fluctuations.

Table 2 illustrates positive correlations across all asset classes. The correlation within the stock market ranges from 0.25 to 0.46. Notably, both gold and oil exhibit relatively strong positive correlations with the US stock market, with correlation coefficients of 0.478 and 0.451, respectively. This suggests that the volatility of these two assets often aligns with movements in the S&P 500. In contrast, certain asset classes display weak positive correlations. Bitcoin, for instance, shows weak positive correlations with all other asset classes (all below 0.26), indicating its relative independence from the fluctuations of other markets. Similarly, the Vietnam market demonstrates weak positive correlations with gold (0.127), oil (0.125) and China (0.144). This may suggest that these markets exert limited influence on the volatility of the Vietnam market due to the relatively low level of integration of the Vietnam stock market with global financial markets.

The analysis first examines volatility spillovers across the full sample period from January 2018 to December 2023. Diagnostic tests are performed before estimation to ensure the data are suitable for the VAR framework. In line with Diebold and Yilmaz (2012), we select a lag length of one for the VAR model based on the Akaike Information Criterion and the Bayesian Information Criterion, and set the forecast horizon to 10 days. The results are reported in Table 3. The diagonal elements of the spillover matrix represent the proportion of forecast error variance in market i explained by its own shocks, whereas the off-diagonal elements capture the spillovers of a shock in market i to another market j. The total directional spillovers from other markets to market i are indicated in the “FROM others” column, while the “TO others” row displays the total directional spillovers from market i to all other markets. The total system-wide volatility spillover index, reported in the lower-right corner of Table 3, indicates that, on average, 20.12% of the forecast error variance across the seven markets is driven by cross-market spillovers.

It is essential to gain a deeper understanding of the variance decomposition structure presented in Table 3. The total directional volatility spillover from one market to another varies significantly. The US stock market has the highest total volatility spillover to other markets (54.18%). The connectedness among developed stocks (the USA and Japan) is due to their high correlation. The Japanese stock market experiences the largest pairwise spillovers from the US stock market (⁠$SJapan ←USH=15.63%$⁠). Despite Japan being a developed market, the impact of its economic shock on other markets is quite limited. Emerging and frontier markets (China and Vietnam) are quite isolated from external shocks. More than 85% of the forecast error variances come from domestic shocks (86.74% for Vietnam and 89.92% for China).

In terms of alternative assets, gold, oil and Bitcoin generally exhibit lower levels of net spillovers compared to developed equity markets, reinforcing their traditional roles as risk diversifiers. Among traditional and modern hedging assets, a shock in the gold market has the strongest impact on other markets (26.85%), followed by Bitcoin (22.55%), and finally oil (14.66%). Though the shock from the Bitcoin market has a significant impact on other markets, it receives little spillover from other markets (external shocks only explain 9.87% of its forecast error variances). Interestingly, shocks from Bitcoin impact these emerging markets more significantly than shocks originating from Japan or other traditional commodity markets. In fact, spillovers from Bitcoin rank just behind those from the US in terms of influence.

Table 3 provides an intuitive assessment of the average volatility spillovers among the assets over the full sample. Although this static analysis yields useful insights into overall spillover patterns, it does not account for the inherently time-varying nature of volatility and may therefore overlook important shifts in spillovers, particularly during periods of heightened market uncertainty. To address this limitation, we employ a rolling-window framework to examine the evolution of volatility spillover effects over time and to capture the dynamic behavior of market interdependencies.

Figure 1 presents the dynamic total spillover index, computed using a 100-day rolling window. The results demonstrate that volatility spillovers are not constant over time but respond strongly to major economic, geopolitical and social events. In general, the index exhibits notable bursts during periods of heightened uncertainty, with significant increases observed during global crises.

On average, approximately 40% of the forecast error variance in 10-day-ahead asset volatility is attributable to shocks from other markets, reflecting moderate but non-negligible interconnectedness. However, this relationship intensifies dramatically during periods of market stress. At the onset of the COVID-19 pandemic and the subsequent global lockdowns in early 2020, the total spillover index surged from around 40% to nearly 75%, indicating a substantial increase in cross-asset volatility transmission. This sharp rise aligns with global investor panic, liquidity shocks and synchronized sell-offs across markets.

After peaking in early 2020, the spillover index declined to around 30% by August of that year, suggesting a temporary easing of systemic pressure. The index remained relatively stable from late 2020 through 2021, with minor fluctuations. However, volatility transmission rose again sharply in early 2022, coinciding with the outbreak of the Russia–Ukraine conflict. This geopolitical shock triggered a renewed surge in uncertainty, particularly through its impact on energy prices, the recovery prospects of Asian markets and the safe-haven demand for gold. Within two months, the total spillover index jumped from below 40%–55%, marking the second-largest spike after the COVID-19 outbreak. Despite this sharp increase, the effects of geopolitical events appear to be more short-lived; the index peaked briefly and then trended downward, stabilizing below 40% by mid-2022.

These dynamics suggest that volatility spillovers tend to intensify during a crisis but moderate relatively quickly once markets adjust to new information. The time-varying analysis strengthens our core argument that while gold, oil and Bitcoin may provide hedging benefits, their safe-haven and diversification properties are not static or unconditional. Their behavior shifts in response to broader market conditions, particularly in crisis periods. These findings reinforce the value of adopting a dynamic spillover framework to assess financial interconnectedness and contribute to the broader literature by offering updated evidence on how traditional and alternative assets behave under extreme stress.

Thus far, we have discussed the total spillover index. This index is interesting, but it ignores directional information. That information is contained in the “FROM others” column and the “TO others” row. We now plot the dynamic gross directional volatilities FROM and TO over time, following the method explained in Section 3.

Figure 2 displays the total directional volatility spillovers from each asset to the rest of the system over time (corresponding to the “TO others” measure in Table 3), while Figure 3 presents the spillovers to each asset from others (the “FROM others” measure). These figures provide a comprehensive picture of how shocks are transmitted and absorbed across assets and markets.

The US stock market consistently stands out as the dominant net transmitter of volatility throughout the sample period. Its outward spillovers intensify sharply during the COVID-19 pandemic, with shocks from the US explaining more than 25% of forecast error variances in other markets. Despite transmitting substantial volatility, the US market remains largely insulated from external shocks, with incoming spillovers averaging around 6%, consistent across both normal and turbulent periods. This reaffirms the US market's dominant role as a transmitter, underscoring its systemic importance in global financial networks.

Gold's role in volatility spillovers is more dynamic and episodic. From late 2018 to the end of 2020, the gold market primarily acted as a net transmitter of shocks. However, between 2020 and late 2021, spanning much of the COVID-19 crisis, it shifted to being a net receiver, reflecting its traditional safe-haven status as investors sought refuge during heightened market uncertainty. Beginning in early 2022, with the onset of the Russia–Ukraine conflict and rising global inflation, gold once again became a net transmitter, with notable impacts on emerging equity markets, particularly China. Gold's spillover index rose to nearly 15%, making it one of the most influential transmitters during that period. These findings suggest that, while gold is traditionally considered a safe haven, its role may reverse during extreme global uncertainty as it becomes a transmitter of volatility. This shift is particularly visible in emerging markets, where gold price movements tend to influence investor sentiment and market behavior more directly, as seen in China's increased gold accumulation in late 2022 amidst real estate and stock market distress.

Crude oil exhibits an inconsistent position, shifting between transmitter and receiver depending on the nature of the shock. Its spillovers tend to rise during geopolitical disruptions, such as the early 2022 energy shock, but generally remain more muted compared to equity markets and gold. This reflects oil's hybrid role as both a commodity and a financial asset, with its spillover behavior shaped by macroeconomic fundamentals and supply-demand imbalances. Oil's transmission role is slightly more pronounced in developed markets, while its spillover impacts on emerging markets remain limited.

Bitcoin exhibits a complex and increasingly important role in the volatility spillover network. While it remains relatively disconnected in calm periods, its net spillovers rise dramatically during stress episodes. From early 2020 to late 2021, we identify two clear periods of significant net volatility spillovers originating from Bitcoin. At the onset of the COVID-19 crisis, Bitcoin acted as a key conduit of volatility, transmitting shocks to equity markets in Japan, China and Vietnam. This highlights its transformation into a systemic player during high uncertainty. Notably, during this time, the US stock market also transmitted shocks to Bitcoin, emphasizing the bidirectional influence between major financial markets and crypto assets. Meanwhile, the Chinese stock market acted primarily as a recipient of Bitcoin-driven shocks, suggesting greater sensitivity to external risk channels. Even during the broad crypto market downturn in 2022, Bitcoin continued to exert a notable influence on other markets, underscoring its growing interconnectedness with traditional financial systems. While its safe-haven status remains questionable, Bitcoin's role as a net volatility transmitter during crises is increasingly comparable to that of the US stock market.

In contrast, China and Vietnam, the two emerging/frontier markets in our sample, consistently exhibit the lowest net outward spillovers, reinforcing their roles as volatility absorbers rather than contributors. This supports our earlier finding that these markets function primarily as net receivers of external shocks, a dynamic that reflects their relatively smaller global footprint or tighter domestic capital controls. This finding highlights differences in volatility dynamics across markets at varying stages of financial development.

These results illustrate that the safe-haven and hedging roles of gold, oil and Bitcoin are neither stable nor universal. Their roles are time-varying and market-dependent. Gold continues to serve as a relative safe haven, particularly in developed markets in normal conditions, but may act as a volatility transmitter during crises. Oil acts as a transmitter when macroeconomic or geopolitical shocks dominate. Once loosely tied to traditional markets, Bitcoin behaves more like a speculative risky asset, occasionally amplifying systemic stress rather than mitigating it. These findings are broadly consistent with prior studies highlighting the conditional nature of safe-haven behavior. For example, Baur and McDermott (2010) and Bouri et al. (2017) show that gold and Bitcoin can offer protection during some crises but not all. Our results further support Shahzad et al. (2020) and Ghorbel and Jeribi (2021), who emphasize the time-varying and crisis-sensitive nature of cross-asset spillovers. However, unlike earlier studies that often focus on a single asset or market pair, our framework allows for a more comprehensive view across multiple asset classes and regions, thereby offering new evidence on how these dynamics differ between developed and emerging markets. Our results diverge from studies that treat Bitcoin as a consistent hedge (e.g. early work by Dyhrberg, 2016) by demonstrating that its spillover effects can rival those of major equity markets during crisis periods. Similarly, while oil is sometimes found to act as a macroeconomic hedge (Sadorsky, 1999), our findings suggest that its hedging function is far less stable, particularly when geopolitical shocks dominate price dynamics.

By jointly analyzing both the directional and net volatility spillovers, our study offers updated empirical support for the argument that no asset consistently provides shelter during systemic shocks and that the effectiveness of risk management strategies depends heavily on timing, market structure and the nature of the crisis. This enhances the understanding of volatility transmission mechanisms and adds a broader, comparative perspective to the existing literature on financial contagion and asset connectedness, highlighting the importance of adopting dynamic portfolio management and risk assessment frameworks in an increasingly interconnected global financial system.

The forecast performance can potentially be affected by assumptions used in estimating the VAR models. To address this, we conduct a sensitivity analysis of the total spillover index by varying the rolling window widths, lag structures and forecast horizons.

We re-estimate the volatility index, employing windows of 75, 100 and 125 rolling days. The trend of the spillover indices (not reported here) is found to be insensitive to the choice of the estimation window length, though the shorter 75-week rolling window generally produces higher spillover indices than the 100 and 125-week rolling windows. This can be attributed to the heightened responsiveness of economic agents to immediate changes in their environment. Additionally, we evaluate the impact of the lag structure by employing lags of 4, 5 and 6. The results indicate minimal sensitivity to the selected lag structure within the spillover index. Lastly, we analyze forecast horizons of 2, 5 and 10 days, and the findings (not reported here) also suggest minimal sensitivity to the chosen forecast horizon.

This paper investigates the dynamic volatility spillovers among gold, crude oil, Bitcoin and equity markets in the USA, Japan, China and Vietnam from 2018 to 2023. By applying the Diebold and Yilmaz (2012) spillover framework to a diverse set of asset classes and markets, we assess the evolving roles of traditional and alternative assets in times of market stress and stability. Our findings reveal three key insights. First, volatility spillovers are time-varying and intensify during major global crises, such as the COVID-19 pandemic and the Russia֪–Ukraine conflict. These episodes highlight the heightened interconnectedness of financial markets during periods of systemic stress and reinforce the need to reassess diversification strategies under such conditions. Second, developed markets, especially the USA, emerge as dominant transmitters of volatility, while emerging and frontier markets such as China and Vietnam are primarily recipients. This asymmetry underscores the structural vulnerabilities of less integrated markets and the global influence of advanced economies. Third, while gold, oil and Bitcoin can offer hedging or diversification benefits during normal periods, their behavior shifts significantly during crises. In particular, gold transitions from a safe haven to a volatility transmitter under intense global uncertainty, oil's hedging role is highly shock-dependent, and Bitcoin, once largely disconnected, now behaves as a key transmitter of volatility, particularly in developed markets.

These findings carry several practical implications. For global investors, the results suggest that traditional assumptions about safe havens and diversification may not hold during periods of extreme uncertainty. Portfolio managers should adopt dynamic risk management strategies that account for the shifting roles of gold, oil and Bitcoin, especially during systemic shocks. For policymakers, the evidence highlights the growing systemic relevance of both traditional and digital assets in transmitting financial stress. Regulatory frameworks in developed markets should recognize the increasing interconnectedness between cryptocurrency markets and traditional financial systems.

For Vietnam, the findings reveal both challenges and opportunities. As a frontier market, Vietnam remains largely a volatility absorber rather than a transmitter, which reflects its relative insulation from global financial turbulence. As Vietnam's capital market liberalizes and attracts more international flows, its sensitivity to external shocks may increase. Portfolio managers in Vietnam should adopt more proactive asset allocation strategies that account for global volatility transmission, rather than relying solely on domestic fundamentals. Vietnamese investors should also exercise caution in assuming that gold or Bitcoin can serve as reliable safe-haven assets during global crises. Instead, dynamic asset allocation and better access to hedging instruments are needed. Vietnam is relatively insulated from global volatility is a double-edged sword. While it implies short-term resilience, it may also reflect a lack of integration that limits liquidity, diversification and market depth. Policymakers should pursue gradual capital market reforms to balance openness with prudential safeguards. Enhancing market transparency, expanding institutional investor participation and improving derivatives markets would strengthen Vietnam's capacity to absorb shocks rather than amplify them. As Vietnam continues to integrate into global capital markets, coordinated macroprudential oversight and improved financial data infrastructure will be critical in managing contagion risks and supporting financial stability.

Looking ahead, several aspects for future research could build on our findings. While this study focuses on gold, oil, cryptocurrencies and stock markets, future work could broaden the scope to include additional asset classes such as bonds, exchange rates or real estate. Including these assets could reveal more intricate interdependencies and transmission mechanisms across financial systems. Methodologically, the VAR-based spillover framework, while effective, may face limitations in high-dimensional settings. Future studies could explore extensions or alternatives, such as factor models, Bayesian networks, or machine learning-driven approaches, to better capture the complexity of today's increasingly interconnected and data-rich investment landscape.