Risks of decentralized finance and their potential negative effects on capital markets: the Terra-Luna case

Abstract

Purpose

This study aims to investigate the impact of the 2022 collapse of the Terra-Luna ecosystem on volatility correlations among digital assets, including U.S. Terra, Luna, Bitcoin, Ether, a Decentralized Finance index and U.S.-sourced conventional assets stocks, bonds, oil, gold and the dollar index. The primary research question addresses whether correlations increased between digital and conventional assets during the collapse.

Design/methodology/approach

A dynamic conditional correlation generalized autoregressive conditional heteroskedasticity model was used to examine changes in volatility correlations during the market crash. Specifically, a data set of 1,442 close prices from 30-minute interval candles of digital and conventional asset prices are considered to provide a granular view of market dynamics during the sample period from January 3rd, 2022, to May 31st, 2022, including the crash event.

Findings

While the dynamic conditional correlation plots of the model indicate increased volatility, the results do not offer sufficient evidence to confirm an increase in correlations between digital and conventional assets during the Terra-Luna downfall. Furthermore, the authors confirm Bitcoin’s role as a diversifier with oil and observe the dollar index maintaining a negative correlation with Bitcoin during the crash, supporting Bitcoin’s function as a hedge against the U.S. dollar. However, the findings during the crash diverge from previous studies, reflecting shifts in correlation patterns in broader market downturns. Specifically, the authors identify the need for adaptive capital allocation strategies, as gold’s oscillation during the period suggests it may not serve as an effective hedge during black swan events.

Practical implications

The findings provide insights for investors, financial institutions and regulators to improve risk management, portfolio diversification, trading strategies and the formulation of consumer protection regulations. In addition, the results underscore the challenges of mitigating risks beyond regulatory measures and emphasize the importance of exercising caution for investors.

Originality/value

This study addresses the research gap in changes between conventional and digital asset volatility correlations during collapses in the digital asset space.

Citation

Santiago, V., Charifzadeh, M. and Herberger, T.A. (2024), "Risks of decentralized finance and their potential negative effects on capital markets: the Terra-Luna case", Studies in Economics and Finance, Vol. ahead-of-print No. ahead-of-print. https://doi.org/10.1108/SEF-02-2024-0075

1. Introduction

Since 2018, the Decentralized Finance (DeFi) space has aimed to create an open, trustless and permissionless financial market, which provides an alternative to centralized financial institutions, third-party players and the traditional financial system. In 2022, markets experienced the downfall of numerous blockchain protocols, projects and ecosystems, largely attributed to inherent vulnerabilities and the fragility of these technologies. One of the most significant failures was the Terra Luna ecosystem, which collapsed alongside its DeFi Protocol Anchor, its native token LUNA and its algorithmic stablecoin U.S. Terra (UST), resulting in a loss of 200bn U.S. dollars (USD) in just 24 h. Understanding the potential impacts of these risks on both digital and conventional asset classes is essential, especially as they remain largely unregulated, leaving DeFi with potentially significant exposure to conventional assets (Carapella et al., 2022). DeFi is still in its early stages and has significant implications for Web 3.0 development and potentially the larger financial system. Reducing such vulnerabilities is essential for the DeFi space to continue making strides in the global economy and foster stability in the DeFi and Web 3.0 ecosystems.

The motivation for our study is inspired by previous research by Geuder et al. (2019) and Aliu (2024), who highlight the recurrent nature of speculative bubbles and sharp collapses and decreases in the crypto economy (e.g., Bitcoin), emphasizing the importance of understanding these dynamics. Furthermore, the works of Antonakakis et al. (2019), Yi et al. (2018), Bouri et al. (2021) and Bajra and Aliu (2023) underscore the importance of closely examining the interactions between conventional and digital assets, particularly in light of their research on the connectedness and spillover effects of crypto assets. Lee et al. (2022) investigate similar factors in the context of the Terra-Luna downfall, concluding that its collapse significantly impacted the connectedness, investor attention and sentiment within the cryptocurrency market. Our study expands the work of Lee et al. (2022) and contributes to the existing literature by addressing the research gap regarding changes in volatility correlations between conventional and digital assets during collapses in the digital asset space. There is no conclusive evidence regarding the relationship between these two domains. In addition, we consider and address the heterogeneous nature of risk spillovers across cryptocurrencies found by Bouri et al. (2023). As one of the most significant collapses to date, the Terra-Luna downfall serves as an ideal case study for investigating the research question of whether the correlation of volatility between digital and conventional assets increased during the crash. Conventional assets are sourced from the U.S. market to ensure robustness and relevance, given its global influence, systemic importance, high liquidity, broad representation and data availability.

The methodology adopted in our research involves a case study approach characterized by the description, understanding, prediction and evaluation of the observed dynamics. This approach can serve as a starting point for inductive theory development and can give valuable insights to relevant stakeholders (Helm et al., 2022). We use a dynamic conditional correlation generalized autoregressive conditional heteroskedasticity (DCC-GARCH) model to analyze changes in volatility correlations during the market crash. The study examined 1,442 30-min interval close prices of digital and conventional assets from January 3rd, 2022, to May 31st, 2022, offering a detailed view of market dynamics during the crash.

Our results contribute to understanding crisis dynamics by exploring previously unexamined correlations among multiple assets. While we observe an upsurge in volatility during the collapse, our findings do not provide conclusive evidence of an increase in correlation between digital and conventional assets during this tumultuous period, with the exception of gold. Most notably, we find that the correlation between bitcoin (BTC) and gold oscillates from a generally negative to positive during a market crisis, building upon the findings of Rudolf et al. (2021) and Aliu et al. (2024). This shift highlights the need for adaptive capital allocation strategies, as gold’s oscillation during the period under review suggests that its role as an effective hedge does not hold during black swan events. Furthermore, we demonstrate that the conclusions drawn by Selmi et al. (2018) and Regaieg et al. (2020), who find that BTC acts as a diversifier for oil and the U.S. dollar, respectively, hold even under extreme market conditions. We find that other cryptocurrencies, equities and bonds maintained a constant correlation with BTC during the collapse.

Our findings serve as a crucial resource for regulators and investors, helping them refine their approaches to risk management, portfolio diversification, trading strategies and consumer protection regulations. Our research is relevant because understanding the dynamics of asset correlations during significant market downfalls, such as the Terra-Luna collapse, is crucial for investors aiming to diversify their portfolios effectively. For investors with portfolios that include conventional and digital assets, or those consisting solely of digital assets, our findings offer important implications for future portfolio management. This research aims to uncover possibly previously unknown correlations in the portfolio asset allocation and help avoid unsystematic portfolio risks. Similarly, financial institutions can develop more effective risk management strategies and refine their investment approaches by understanding how these assets interact during heightened volatility. For policymakers and regulatory bodies, insights into the volatility correlation between these asset classes can inform the development of appropriate regulatory frameworks. This understanding can inform decisions on stress testing, capital requirements and the overall stability of the financial system.

The remainder of this study is organized as follows. Section 2 presents the theoretical background of the study and provides a review of existing literature. Section 3 explains the research design and the statistical model applied. Section 4 offers the empirical findings and a discourse on the key research discoveries. Finally, in Section 5, we conclude and underscore the implications of this investigation. In addition, we outline the study’s limitations and propose suggestions for potential avenues for future research.

2. Literature review

While concerns regarding collateralized stablecoins primarily lie in the difficulty of valuing reserves and the absence of mandatory disclosure, algorithmic stablecoins circumvent such concerns by maintaining an algorithmic peg on the premise that their value can be upheld through controlling the supply of its ecosystem token, whose value is by user expectations of the future purchasing power rather than traditional collateral (Bullmann et al., 2019; European Central Bank, 2019; Salami, 2021). The concept of a pure algorithmic stablecoin, as demonstrated by Terra-Luna, has faced significant scrutiny, as prior attempts by NuBits and IRON Finance led to similar collapses (Clements, 2021; European Securities and Markets Authority, 2022; Phan, 2022; Reserve Research Team, 2018; Reynolds, 2021). The idea that a native token such as LUNA can absorb volatility shocks holds under normal market conditions. In extreme cases, however, without collateral or intrinsic value, a loss of confidence in expected future returns causes hyperinflation (European Securities and Markets Authority, 2022). Consequently, the European Central Bank attests that algorithmic stablecoins remain a theoretical alternative rather than a practical solution (European Central Bank, 2019). Terra-Luna attempted to address the issue of intrinsic value by leveraging network effects and allocating funds accrued through transaction fees for ecosystem development in their two-coin system (Kereiakes et al., 2019). Despite this, in April 2022, Coindesk and Swissborg reported the risk of a bank run on Terra-Luna’s two-coin system (Bonato, 2022; Morris, 2022). A month later, similar to its algorithmic predecessors, when UST’s demand experienced a shock, hyperinflation of its native token broke its peg, causing a collapse in market capitalization and increasing the system’s fragility to the point of failure (Buterin, 2022; Phan, 2022; Reserve Research Team, 2018; Reynolds, 2021).

While algorithmic stablecoins have had limited success, MakerDAO’s pseudo-algorithmic DAI coin has shown resilience during periods of significant collateral volatility, thanks to its decentralized seigniorage base and overcollateralized, diversified on chain backing (Bullmann et al., 2019; Buterin, 2022; European Central Bank, 2019; MakerDAO, 2020). Other solutions have been proposed, such as third-party institutions creating stable infrastructure through a central bank digital currency or granting full transparency to issuer balance sheets. However, both solutions compromise the unique selling points of DeFi by limiting privacy and decentralization (Chiu et al., 2022). While some view digital assets as an ongoing experiment, regulatory policies must be carefully considered before systemic risks emerge, especially as the frequency and impact of such collapses continue to grow (Carapella et al., 2022; Cermak, 2017; European Securities and Markets Authority, 2022). However, little progress has been made in setting any global regulatory standards (Salami, 2021). As a first step, in April 2023, the Members of the European Union Parliament approved the Markets in Crypto-Assets Regulation (MiCA), which sets standards for the issuance, disclosure, collateral and authorization of such tokens, taking into account the significance of a given stablecoin and its interconnectedness with the traditional financial system (EU Parliament, 2023; General Secretariat of the Council of the Council, The European Union, 2019; Salami, 2021).

Interdependencies between the cryptocurrency and traditional financial markets allow for the transmission of shocks between them (European Securities and Markets Authority, 2022). Links between the two systems often lead to material value changes in one market spilling over into the other, such as in the case of assets as collateral for liabilities. BTC is more volatile even in regular times and poses a downside risk spillover to conventional assets, underscoring the need for careful monitoring to ensure financial stability (Smales, 2019). The literature reveals ongoing debates about BTC’s role as a hedge, haven or diversifier against conventional financial assets (Aliu et al., 2024). Studies by Rudolf et al. (2021), Stensås et al. (2019) and Yi et al. (2018) suggest that BTC can serve as a diversifier or hedge for investors, while Aliu et al. (2021) highlight how the positive correlation among cryptocurrencies limits this potential. While Yi et al. (2018) find that cryptocurrencies differ from traditional financial assets such as stocks in large, others find that cryptocurrencies, particularly BTC, tend to be associated with tech stocks like PayPal and CashApp, as well as the NASDAQ (Falk and Hammer, 2022). This is further supported by Bajra and Aliu (2023), who find BTC most notably impacts speculative assets, such as equity indices, while having less influence on less speculative assets. During the Terra-Luna crash, Arcane Research reported all-time highs in the 90-day correlation between BTC and the S&P 500 (Falk and Hammer, 2022). This finding is consistent with the International Monetary Fund (IMF) report, which shows that BTC has become increasingly correlated to the S&P500 since 2020, with BTC spillovers in both directions being of similar magnitude (Adrian et al., 2022). Further, while BTC is often perceived as more volatile than stocks, Hajric (2022) observed that this was not the case during the low-volume markets of October 2022.

BTC exhibits similarities to both gold and the U.S. dollar in terms of volatility and shock influence (Dyhrberg, 2016). Regaieg et al. (2020) suggest that BTC exhibits characteristics of a diversifier for currencies, particularly as a medium of exchange. Further, they find that BTC’s use as a medium of exchange influenced demand for its returns more than temporary price shocks. Research on the risk management capabilities of gold as a hedge against the dollar may extend to suggest that BTC could also offer some risk management capabilities against the dollar (Capie et al., 2005; Dyhrberg, 2016). However, recent studies by Aliu et al. (2024) have been inconclusive thus far. BTC and gold are highly correlated and exhibit positive short-term spillovers on each other, supported by Rudolf et al. (2021), who find that BTC is increasingly being held as a store of value rather than as a medium of exchange (Aliu et al., 2024). This trend became particularly evident during the COVID-19 pandemic when BTC’s perception as a hedge against the economic recession decreased its volatility and increased its returns (Rudolf et al., 2021). Some studies suggest that BTC outperforms both gold and major stock indexes as a hedge and investment, exhibiting superior returns (Bouri et al., 2020; Jareño et al., 2020). Unlike gold, whose value is partly based on intrinsic worth, BTC’s value is entirely driven by market dynamics, making its variance more influenced by external forces. This characteristic makes BTC an attractive option for diversifying portfolios (Dyhrberg, 2016; Hatemi-J et al., 2021; Tully and Lucey, 2007). In 2018, Selmi et al. found that BTC and gold serve as a hedge, safe haven and diversifier for oil, with portfolios containing both assets outperforming under different market conditions. While much of the literature focuses on BTC’s relationship with conventional assets, Yi et al. (2018) suggest that investors and portfolio managers should consider a range of cryptocurrencies rather than relying solely on BTC. Wang et al. (2022) find that the dynamic correlation coefficients of BTC and U.S. Treasury bonds fluctuate near zero; however, they exhibit significantly stronger linkage during extreme market events. This finding is critical for assessing systemic importance, as U.S. Treasury bonds are often seen as a proxy for global markets due to their high liquidity and low risk relative to other bonds or asset classes (Habib et al., 2020; He et al., 2016).

When turning to the relationships among digital assets, Aliu et al. (2021) demonstrate that the heightened positive correlations and extreme volatility within cryptocurrencies significantly diminish their diversification benefits, highlighting the complex interplay between digital assets. Furthermore, Bouri et al. (2021), Antonakakis et al. (2019) and Yi et al. (2018) all find that the connectedness of cryptocurrencies increases during extreme market events and times of high volatility. In Bouri et al. (2023), a study on dynamic lower tail dependence, the authors find that risk spillover effects vary significantly among different cryptocurrencies, further motivating subcategorization in this study. Conversely, Baur and Hoang (2023) find spillovers amongst cryptocurrencies are generally stable in crisis and play a small role relative to correlations that increase with financial contagion. Their emphasis on correlations over spillovers is credited to the transparency of the space, which limits asymmetric information and, thus, the possibility for spillovers caused by market inefficiencies. These findings support the possibility of a change in correlation between digital and conventional assets during the Terra-Luna crash, validating the purpose of this study. Furthermore, this has implications for hedging strategies and risk management for investors (Ji et al., 2018).

Research indicates that BTC and other highly market-capitalized currencies do not solely influence the cryptocurrency market. This finding is supported by the Granger causality approach applied by Bouri et al. (2019), the threshold weighted-minimum dominating set methodology used by Papadimitriou et al. (2020) and a “pure” vector autoregression spillover study by Baur and Hoang (2023), who attribute small spillovers amongst cryptocurrencies to investors having a generally homogeneous attitude toward them. While both their findings state that BTC, as the largest coin by market capitalization, is not the clear leader regarding volatility connectedness, Yi et al. (2018) suggest that connectedness among cryptocurrencies does not necessarily depend on their size. This supports the findings of Bouri et al. (2019, 2021) and suggests that similar dynamics may have occurred with LUNA and UST during the collapse, as observed by Antonakakis et al. (2019). Notably, Bouri et al. (2021) emphasize the importance of incorporating various leading cryptocurrencies into any monitoring procedures related to the stability of network connectedness. Other scenarios include the possibility that BTC is losing dominance or that the convergence of cryptocurrencies is causing them to become increasingly homogenous in the eyes of investors (Papadimitriou et al., 2020; Ji et al., 2018).

In conclusion, the literature widely agrees that the inherent vulnerabilities of algorithmic and uncollateralized DeFi protocols pose a fundamental challenge to their stability. Notably, this highlights the practical limitations of pure algorithmic stablecoins. While approval of standards and regulations such as MiCA is affirming, they come at an untimely cost, and their effectiveness is yet to be proven. Regarding their suitability as investment instruments and diversification tools for portfolios, existing studies offer conflicting perspectives on digital assets. While some research findings suggest that they can serve as effective hedges for conventional assets, others posit that they are a reliable store of value. Given the interconnected nature of digital assets in terms of volatility, as previous studies have indicated, this information is paramount to investors seeking viable investment options and diversification strategies. However, events such as the Terra-Luna breakdown continue to unsettle investors. Consequently, this research seeks to determine whether there is empirical evidence of an increase in the correlation of volatility between digital and conventional assets during the Terra-Luna crash. Such evidence would have profound implications for the role of digital assets as investment instruments and diversifiers in portfolios. While the existing body of literature has already examined the weaknesses inherent in DeFi architecture and has explored the potential spillover effects between digital assets and traditional financial investments, there is no conclusive evidence regarding the relationship between these two domains. Furthermore, given that the Terra-Luna case is a recent and highly impactful event within the financial markets, no research study has investigated it and its repercussions on traditional financial markets.

3. Research design and methodology

The variables included in our model were chosen to provide a comprehensive sample of traditional and digital asset classes. The sample of conventional assets was sourced from the U.S. market due to its global influence, systemic importance, high liquidity, broad representation as a global market and data availability. BTC is taken as the primary sample of the cryptocurrency market due to its dominant market capitalization and ease of comparability to previous research. Ether (ETH) is considered due to its prominence in the Web 3.0 and DeFi space, as well as lacking affiliations from the crash, providing robustness as BTC may have been manipulated during the sample period leading up to the crash (Briola et al., 2022; Falk and Hammer, 2022; Lee et al., 2022). These considerations, together with Bouri et al.’s (2021) findings stressing the importance of considering various leading cryptocurrencies during extreme market conditions, led to the inclusion of both BTC and ETH to provide reference and robustness in the DCC analysis. A DeFi index is included to investigate the crash’s impact on the originating DeFi ecosystem. Finally, LUNA and UST were included as the two main assets involved in the Terra-Luna crash [1].

As no comprehensive DeFi index encompassing the sample period currently exists, a market capitalization-weighted DeFi index was constructed based on the following criteria. DeFi projects with the highest total value locked (TVL) as of Jan 1, 2022, the sample start date, were selected. Projects built on the Terra-Luna Blockchain, such as Anchor Protocol, were excluded due to their direct role in the crash. Projects were included until the TVL of the most minor project fell below 10% of the most significant project. To prevent the dominance of any single category (e.g., lending), the index limits dominance to 50% per category. To ensure tokens are liquid and provide a representative sample, included assets must represent tokens that serve as a bearer instrument. As a result, this excludes wrapped, synthetic tokens derived from other assets, tied to physical assets, or representing claims on other tokens. Further breakdown of constituents can be found in Table 1.

Returns of 30-min candles retrieved from Trading View were calculated to identify critical points of the crash. Due to data limitations of novel assets, consistency was prioritized in data collection. Contract for Difference products quoted by Trading View were used to represent the underlying value of speculative traditional conventional assets to maintain consistency. Bond value was expressed as yield, reflecting their stronger association with capital preservation than speculative investments. Due to the novel and niche nature of digital assets, price data was quoted by multiple exchanges, prioritizing those with the most extensive availability. Further details can be found below in Table 2.

Missing data points were estimated by linear interpolation for the constituents of the DeFi Index and the U.S. 10-year treasury bond (US10Y). Digital asset prices were retrieved in U.S. dollar Terra (USDT) and converted to USD using the 30-min USDT/USD exchange rate.

A sample period spanning from January 3rd, 2022, 14:30 UTC, to May 31st, 2022, 23:00 UTC, comprised 1,442 observations taken during the operational hours of the U.S. stock market. This is the broadest frame available as prior data for LUNA and UST. Critical times in the sample include the start and end of the crash period, which are reflected in the sample from May 9, 2022, 14:30 UTC to May 16, 2022, 14:30 UTC, to include the reopening of conventional asset markets following the collapse, which bottomed May 13, 2022, 03:00 UTC.

To address the empirical research question, a multivariate DCC-GARCH model was used to examine the asymmetric effects of dynamic correlations of volatility and overall volatility correlations between digital and conventional assets during the crash period, as applied by Batten et al. (2023) and Kinateder and Papavassiliou (2019). Volatility is prioritized in this study due to its ability to capture the market’s overall uncertainty and risk perception. The crypto space has a high participation of nonprofessional investors, who often react to dramatic swings in market sentiment and perceive risk more strongly, making volatility a more comprehensive metric for understanding the market than return connectedness and having a more immediate effect on metrics such as value at risk (VaR) and maximum drawdown (Batten et al., 2023).

The selection of the DCC-GARCH model is based on its superior capacity to capture time-varying correlations between multiple financial time series, which is particularly critical in the context of this study (Engle, 2002; Engle and Sheppard, 2001). During periods of market turmoil, such as the Terra-Luna collapse, correlations between assets can shift rapidly and unpredictably (Engle, 2002). The DCC-GARCH model’s ability to dynamically adjust and directly model these conditional correlations ensures a precise analysis of how the relationships between digital and conventional assets evolve over time (Engle, 2002; Engle and Sheppard, 2001). While popular models such as the Copula-GARCH model are adequate for modeling joint distributions and tail dependencies, they focus more on dependency structures rather than dynamic correlations, making them less suitable for analyzing the shifting correlation structures that are crucial for understanding market dynamics during a crisis (Jondeau and Rockinger, 2006). Consequently, the DCC-GARCH model provides a more robust and contextually appropriate framework for this study (Engle, 2002).

Furthermore, the model was selected for its ability to analyze dynamic correlations and various effects between multiple time series, overcoming the limitations of traditional volatility models such as standard or moving average deviation. These models often fail to account for the heightened volatility of variance covariances of returns during economic collapses such as that in focus in this paper (Rudolf et al., 2021). A high dcca1 coefficient indicates past dynamic correlations have a significant effect on current correlations, while a low dccb1 coefficient indicates current and past errors have minimal impact on current correlations (Bauwens et al., 2006; Engle, 2002; Rudolf et al., 2021).

4. Results and discussion

In the following, we describe the findings of the DCC plots created using the multivariate DCC-GARCH model about the volatility correlations for the entire sample, as shown in Table 3. As the focus of this paper is on the relationship between digital and conventional assets, additional volatility correlations that can be derived from the model will not be addressed, particularly between conventional assets and digital assets, other than the DeFi index [2].

As shown in Figure 1, the S&P500 and BTC maintain relatively stable positive DCCs in a corridor between 44% and 67.27%, both before and during the crash of Terra-Luna. We do not see any significant deviation from the positive correlation at moderate strength of 53.67% for the full sample in Table 3. This stability is explained by the IMF’s statement that BTC has become increasingly correlated to the S&P500 since 2020, with BTC spillovers in both directions being of similar magnitude (Adrian et al., 2022).

US10Y and BTC are positively correlated with a weak strength of 20.87% during the sample period. Figure 1 indicates volatile DCCs in the U.S. 10-Year Treasury Bond – Bitcoin correlation, ranging from 3.66% to 42.55%. However, we do not observe any significant change during the Terra-Luna crash. Gold and BTC are negatively correlated with a weak strength of −12.00% during the sample period. Figure 1 indicates that the DCCs regularly oscillate between positive and negative, peaking at 11.73% during the Terra-Luna crash, with previous limits at 4.10% and −47.03% during the sample. This indicates a significant change during the crash period, as it marks a new high in correlation between the two assets. The DCC of BTC and gold is the only pairing to exhibit a directional change in DCC when breaking out of its corridor, becoming uncharacteristically positively correlated with BTC during the crash period. This directional change to a positive correlation occurred only once prior in the sample when reaching the corridor high of 4.10%, with the DCC of all other points being less than or equal to zero. Gold is the only asset that broke out of its DCC corridor during the crash period. Oil and BTC are positively correlated with a weak strength of 2.96%, the weakest of conventional assets compared to BTC. Like gold, the DCCs in Figure 2 further indicate that this correlation periodically experiences large periods of volatility. Unlike gold, however, oil is the only asset to regularly oscillate between positive and negative DCC values with BTC, ranging between 30.94% and −26.38% over the full sample period. Despite an increase in the correlation peaking at 26%, oil does not break the conditional correlation corridor mentioned above, indicating no significant change during the crash period. U.S. Dollar Index (DXY) and BTC have a negative correlation with a weak strength of −20.87% during the sample period. While Figure 2 indicates occasionally volatile DCCs in a corridor from −47.83% to 2.84%, we do not see any significant change during the crash of Terra-Luna. While DXY and gold are negatively correlated to BTC throughout the sample, DXY is the only asset to maintain this negative correlation during the crash. It should be noted that the DCC between DXY and BTC exhibited a weakly positive correlation of 2.84% on March 28, 2022, but this falls outside the crash period. The DeFi Index and BTC have a positive correlation with a strength of 33.95% during the sample period. While Figure 2 occasionally indicates significant increases in DCCs by up to 30%, with a corridor between 63.42% and 14.69%, we do not see any change during the crash of Terra-Luna, which would significantly differentiate it from other increases in the conditional correlation.

The results mentioned above were substantiated through a robustness analysis using the DCCs of ETH instead of BTC for each asset. Analyzing the volatility correlations between ETH and the S&P500, US10Y, gold, oil, DXY and the DeFi Index, respectively, shows similar results.

In addition, while not all assets correlate with LUNA during the entire sample, this changes during the crash period. LUNA reaches a new high in DCC with all variables considered, except for bonds. Figure 3 shows the DCC for each variable, with LUNA referenced below in further detail.

The S&P500 saw very high volatility in DCCs with LUNA during the crash period. It reached a new low of 11.24% from its precrash corridor and significantly below the positive moderate correlation of 33.58%. While US10Y experiences some volatility with LUNA during the crash period, it is the only variable that does not break out from its precrash DCC corridor, which ranges from 35.17% to −0.031%, with a positive correlation with a weak strength of 13.63% for the full sample period. This aligns with the literature review, as US10Y is considered the most stable asset class among those selected (Habib et al., 2020; He et al., 2016). It should be noted that US10Y still experienced heightened volatility during the crash, consistent with the findings of Wang et al. (2022) that the link between BTC and bonds increases during extreme events. Despite this, the largest increase in DCC did not mark one of the three highest levels during the sample period. Gold, oil, DXY and the DeFi Index all break out of their DCC corridors, with LUNA reaching new highs in positive DCC of 26.9%, 32.55%, 7.89% and 53.55%, respectively, significantly above their respective correlations of −1.12%, 2.37%, −15.48% and 22.30% for the full sample as seen in Table 3. In addition, oil and the DeFi Index show new lows in the later days of the crash period, of −31.77% and −16.49%, respectively.

Finally, it is worth noting that the DCC between BTC and LUNA, as well as the DCC between ETH and LUNA, reach new highs of 73.81% and 75.11%, respectively, which can be seen in Figure 4, up from a correlation of 54.18% and 54.68% for the entire sample. This suggests that digital assets experienced a significant increase in DCC with LUNA during the crash period. In addition, ETH’s DCC with LUNA reached a new low of 28.76% during the crash. The increase in DCC among digital assets is in line with the literature of Bouri et al. (2021), Antonakakis et al. (2019) and Yi et al. (2018), who find increased connectedness amongst cryptocurrencies during extreme market events and periods of high volatility.

The dcca1 coefficient of 0.009 shown in Table 4 suggests that the past dynamic correlations have a weak influence on the current correlations, indicating that the variables exhibit relatively independent dynamics or that the relationships between them are not persistent over time. However, as the p-value of the dcca1 parameter shown in Table 3 is 18.37%, significantly outside the 95% confidence interval, a conditional relationship among the variables included in the model is likely. This indicates that the autoregressive component of the DCC is not statistically significant and reinforces that all variables may not be related to one another in the short term, as past dynamic correlations between the variables do not strongly influence the current dynamic correlations. Given the substantial changes observed in the DCC pairings involving LUNA during the crash period, an inconsistency in the relationships between the variables over time is plausible. Nonetheless, further examination of all pairings and additional research methodologies are required to confirm or refute this hypothesis.

The dccb1 coefficient of 0.688 indicates that current and past errors have a moderate to strong influence on the current dynamic correlations, suggesting the existence of lagged effects of the variables on each other or that the relationships between the variables are persistent over time. The p-value for the dccb1 is highly significant. This indicates that all variables are related to one another in the long run, as the past dynamic correlations influence the current dynamic correlations between the variables. This also explains why an increase in correlation may not be apparent in the plots. In conclusion, we do not see an increase in the correlation of volatility between digital and conventional assets during the Terra-Luna crash.

The alpha and beta coefficients for the univariate GARCH models used as inputs to the multivariate DCC-GARCH model above are presented in Table 5. The alpha1 coefficients of all variables suggest that the past variance weakly influences the current variance, with values less than 7%. In addition, all univariate GARCH coefficients, apart from LUNA, are statistically significant at both the 95% and 99% significance level. LUNA’s alpha coefficient being 10% could be explained by the crash period included in the sample, as the probability that the estimated value of the alpha parameter occurred by chance is relatively high, considering the black swan event of the crash is unlikely to be replicated in a sample of another period.

These alpha coefficients could potentially influence the dynamic correlations between the variables and, therefore, explain the high significance of the dcca1 term mentioned above. Note that the dcca1 parameter and the alpha parameters are not directly related, as they are calculated and used in different contexts. The beta coefficients for the univariate GARCH models used as inputs to the multivariate DCC-GARCH model range between 91.20% and 99.53%, suggesting that current and past errors strongly influence the current variance. This could indicate that there are lagged effects of the errors on the variance or that the relationships between the errors and the variance are persistent over time. The standard error of the alpha1 and beta1 coefficients of UST and LUNA exhibit higher magnitudes compared to the other assets of the univariate GARCH model. These higher errors reflect the significant decline and extreme volatility that occurred during the collapse of Terra-Luna.

In summary, the empirical results do not provide sufficient evidence to confirm a rise in the correlation between digital and conventional assets during the Terra-Luna collapse. These findings offer investors and financial institutions insights to improve risk management, portfolio diversification and trading strategies. In addition, they may provide guidance for regulators in the formulation of consumer protection regulations.

By understanding the co-movements of digital and conventional assets during periods of increased volatility, decision makers can refine their decision-making processes, enabling more informed risk management strategies and potentially minimizing financial losses. In particular, our research findings align with those of Selmi et al. (2018), shedding light on the distinct qualities of BTC as a diversifier when paired with oil. Throughout our sample period and the subsequent crash period, BTC consistently exhibits a weak correlation that oscillates near zero. This resilience in correlation dynamics underscores BTC’s potential as a diversifying asset, aligning with the observations made by Selmi et al. (2018). Moreover, our study delves into the relationship between BTC and the Dollar Index (DXY), revealing a noteworthy aspect of BTC’s role as a hedge. In contrast to other assets, DXY is identified as the sole asset that achieves and maintains a negative correlation with BTC without succumbing to significant impact during market crashes. This finding aligns with the conclusions drawn by Regaieg et al. (2020), emphasizing BTC’s potential as a hedge against currency-related risks.

Despite these promising attributes, our results recommend a nuanced approach for investors contemplating portfolio diversification with BTC and gold. This caution is particularly relevant in scenarios with a directional shift to a new high in positive DCC during market crashes. Understanding the intricacies of these correlations is crucial for optimizing portfolio resilience and performance. As long-term investment strategies rely on stable correlations between assets, recognizing how these correlations shift during crises allows for more resilient strategic asset allocation. While BTC maintains a positive correlation with other digital and conventional assets throughout our sample, the absence of significant deviation during the crash period prompts a nuanced interpretation. It becomes evident that BTC, while potentially serving as a diversifier and hedge in specific contexts, may not consistently fulfill this role against broader market indices such as the S&P 500, bonds or the DeFi Index. This nuanced perspective underscores the complexity of BTC’s role in diversified portfolios and necessitates a comprehensive understanding of its dynamic correlations with various asset classes.

5. Conclusion

This study investigates the relationships between the volatility of digital and conventional assets. To achieve this objective, we leverage a prominent and noteworthy event within the financial markets – the Terra-Luna collapse, one of the most remarkable occurrences in recent financial history. Our results contribute to understanding crisis dynamics by exploring previously unexamined correlations among multiple assets.

While the DCC plots of the multivariate DCC-GARCH model indicate increased volatility during the Terra-Luna crash, our empirical results do not reveal an increase in the volatility correlation between digital and conventional assets during this period. The results offer valuable information on the market dynamics during periods of crisis, as DCC correlations between multiple digital and conventional assets were previously unexplored. Specifically, our findings support those of Selmi et al. (2018), as BTC exhibits qualities of a diversifier with oil. Further, we identify DXY as the only asset that maintains a negative correlation with BTC without significant impact during the crash, signifying BTC’s role as a hedge for DXY, confirming the findings of Regaieg et al. (2020). However, our results deviate during the crash period from the generally negative correlation found by Rudolf et al. (2021), in line with the directional change in correlation observed in downward spirals in higher time frame by Aliu et al. (2024) during COVID-19. This aspect of asset correlation resilience has significant implications for portfolio hedging strategies, and investors should be cautious in diversifying portfolios with BTC and gold to avoid overexposure to market crashes. Effective hedging strategies require accurate estimates of return correlations for hedged assets. If the nature of these correlations is highly dynamic, the hedge ratio and other parameters may need to be adjusted to account for recent volatility (Engle, 2002). Throughout our sample period, BTC maintains a positive correlation with other digital and conventional assets without significant deviation during the crash. However, BTC cannot be seen as an effective diversifier or hedge to the S&P500, bonds or the DeFi Index.

Our results have practical implications for investors, policymakers, and financial institutions. Firstly, understanding the dynamics of asset correlations during significant market downturns, such as the Terra-Luna collapse, is crucial for investors aiming to diversify their portfolios effectively. At present, the high volatility of crypto assets, coupled with the absence of sufficient regulation and the instances of fraud scandals associated with cryptocurrency exchanges, continue to pose significant risks for investors (Aliu et al., 2024). By comprehending how digital and conventional assets behave in tandem during times of heightened volatility, investors can make more informed decisions to manage risk and potentially mitigate losses.

Secondly, financial institutions can benefit from exploring the correlation between digital and conventional assets during significant market disruptions. By understanding how these assets interact during periods of heightened volatility, institutions can develop more effective risk management strategies and refine their investment approaches. This understanding can enhance their ability to navigate turbulent market conditions and bolster their resilience against unforeseen market shocks.

Specifically, the aforementioned parties benefit from our findings as they confirm that the absence of increased correlation among diverse assets during volatile periods sustains diversification benefits, which are crucial for portfolio stability. For example, low correlation among selected digital assets could have mitigated losses and stabilized portfolios during the Terra-Luna crisis. If correlations had increased, metrics such as VaR, maximum drawdown and portfolio variance would be significantly impacted, leading to an underestimation of risk and inadequate capital allocation. Most notably, gold’s shift from a negative to positive correlation during the period indicates its role as an effective hedge, though this does not hold during black swan events in the space. This implies the need for additional capital allocation strategies and dynamic models of financial institutions and portfolio managers.

Thirdly, for policymakers and regulatory bodies, insights into the volatility correlation between these asset classes can inform decisions on stress testing, capital requirements, and ultimately, the development of appropriate regulatory frameworks. While some investors may view digital assets as an ongoing experiment, regulators must proactively address systemic risks before they potentially materialize, especially considering the increasing frequency and impact of such collapses. As the intersection between digital and traditional finance continues to evolve, regulators must understand the interconnected risks and potential systemic implications that may arise during market upheavals. Therefore, our results can aid in designing more robust regulatory measures that ensure financial stability and safeguard investor interests.

In conclusion, looking at the inherent risks within the Terra-Luna ecosystem, solutions for minimizing the risks of collapses akin to those of Terra-Luna lie primarily in regulation and investor prudence. These outcomes underscore the challenge of mitigating risks beyond the purview of regulation and highlight the importance of investor caution. Moreover, Terra-Luna was not the first algorithmic stablecoin based on seigniorage to collapse, and the associated risks were previously flagged by SwissBorg and Coindesk, as noted by Bonato (2022) and Morris (2022). Given the limited alternatives available, stakeholders should be wary of algorithmic stablecoin systems and DeFi projects. Simultaneously, regulators should swiftly and seriously consider proposals such as MiCA to avert such events in the future.

Our results are subject to certain limitations, which, at the same time, pave the way for future research. The primary constraint of this study lies in the limited availability of data for certain assets before Jan 1, 2022. A broader study could circumvent this by considering digital assets on a rolling basis by market capitalization over time, allowing for consideration of multiple collapses and identifying trends. This would enable regulators and investors to forecast and identify thresholds for stability, considering the exposure of conventional assets to digital assets. A key consideration in interpreting the model is defining a significant change in the DCC, which is particularly relevant in determining the rejection region of the null hypothesis. For this study, a change in DCC is considered significant if it broke out of the corridor defined by the limits established in the sample period before the crash. A more extended sample would likely expand this corridor as other extreme market events are encompassed, leading to different results.

To better understand market dynamics, prospective studies would be keen to investigate the flow of funds across both digital and conventional asset classes during collapses in the DeFi space. Considering novel variables and the public availability of information, such as changes in TVL across major DeFi protocols and tracking wallet activity, provides a streamlined approach to monitoring the flow of funds within, as well as entering and exiting, the space. Furthermore, since the collateral for protocols in the DeFi space consists not only of their native token but various types of digital assets, changes in TVL provide a more accurate gauge of investor sentiment than returns.

Scholarly research on digital assets still has a notable gap, hindering the development of regulatory framework proposals with robust solutions for minimizing the risks of such collapses. Incorporating subcategories about the diverse services offered by DeFi and integrating Non-Fungible Tokens (NFTs) would address this gap and provide a more comprehensive and nuanced understanding of the interrelationships and dynamics of digital assets. Such findings hold significant implications for institutions and investors in risk management, portfolio diversification, and trading strategy.

Additional research encompassing multiple collapses over a broader period would provide insights into the exposure of investors, institutions, and ecosystem participants relative to similar collapses in the DeFi space. This expanded scope would enable the identification of trends and establish benchmarks for market stability, consequently facilitating the development of policies to safeguard investors and enhance systemic stability.

Figures

Figure 1.

S&P 500, U.S. 10 Year treasury bond and Gold – Bitcoin conditional correlations

Figure 2.

Oil, U.S. Dollar index and DeFi Index – Bitcoin conditional correlations

Figure 3.

Full sample conditional correlations between LUNA and modeled assets

Figure 4.

Full sample bitcoin – LUNA conditional correlation plots

Table 1.

DeFi Index constituents

Asset		Category					Jan 1, 2022 TVL
Curve		DEX						23.25
Maker		CDP						17.5
Aave		Decentralized money market with lending and governance						14.21
Compound		Lending						8.9
Uniswap		DEX						8.36
PancakeSwap		DEX						5.53
Yearn finance		Yield aggregator						4.12
SushiSwap		DEX						3.91

Notes:

TVL and DeFi function category for constituents of the DeFi Index composed for evaluation of DeFi sectors exposure to the Terra Luna crash. The index is weighted by projects TVL as of 1 January 2022, providing an ideal representation during the start of the sample period. To ensure robustness, projects built on the Terra-Luna Blockchain, such as Anchor Protocol, were excluded due to their direct role in the crash. DEX stands for Decentralized Exchange, and CDP for Collateralized Debt Position. A cap of 50% dominance per category was used to prevent overrepresentation of a particular DeFi function, ensuring a liquid and representative sample

Sources: Authors’ own work; data from Trading View

Table 2.

Asset descriptives

Asset	Asset category	Used as	Price calculated by
S&P 500	Traditional index	Dependent variable	SPX
US 10 Year Treasury Bond	Traditional	Dependent variable	Trading view
DXY dollar index	Traditional index	Dependent variable	Trading view
Gold	Traditional	Dependent variable	Trading view
Brent crude oil	Traditional	Dependent variable	Trading view
USD Tether	Digital (stablecoin)	Conversion factor	Binance
US Terra	Digital (stablecoin)	Dependent variable	Huobi
Luna	Digital	Dependent variable	Huobi
Ether	Digital	Dependent variable	Huobi
Bitcoin	Digital	Dependent variable	Huobi
DeFi index (components)	Digital index	Dependent variable
Curve	Digital	DeFi index component	Kucoin
Maker	Digital	DeFi index component	Huobi
Aave	Digital	DeFi index component	Huobi
Compound	Digital	DeFi index component	Huobi
UniSwap	Digital	DeFi index component	Huobi
Cake	Digital	DeFi index component	Kucoin
Yearn	Digital	DeFi index component	Kucoin
SushiSwap	Digital	DeFi index component	Huobi

Notes:

Overview of variable descriptives sampled to evaluate the effect of the Terra Luna crash on digital and conventional asset classes. Details on the use of each asset are included to provide transparency regarding their role in the DCC-GARCH Model applied. These details include asset categories, data use and price calculation method applied before being entered into the model to display the consistency and reliability applied across the sample period from January 3, 2022, to May 31, 2022. The components of the DeFi index used are indented and highlighted in italics

Source: Authors’ own work

Table 3.

Traditional and digital asset return correlation

Asset	SP500	US 10Y	GOLD	OIL	DXY	BTC	ETH	DEFI INDEX	UST
SP500	1
US 10Y	0.156	1
GOLD	−0.113	−0.236	1
OIL	0.021	0.011	0.403	1
DXY	−0.358	−0.005	−0.077	0.073	1
BTC	0.537	0.209	−0.12	0.03	−0.231	1
ETH	0.551	0.228	−0.091	0.035	−0.225	0.923	1
DEFI INDEX	0.162	−0.078	−0.056	−0.047	−0.083	0.34	0.426	1
UST	0.009	0.009	−0.041	−0.069	0.058	0.054	0.065	−0.008	1
LUNA	0.336	0.136	−0.011	0.024	−0.155	0.542	0.547	0.223	0.423

Notes:

Return correlations between traditional and digital assets during the Terra Luna crash, represented by a sample period from January 1st through May 31st 2022. Asset classes included represent the broader macroeconomic environment to determine relationship beyond the scope of the crypto space. This table highlights the correlation strengths observed over the full sample period, providing insights into the dynamic relationships between these asset classes prior to and during the Terra-Luna crash. The table is referenced later to draw assumptions about changes in dynamics during the crash period further evaluated by the DCC-GARCH Model of the study

Source: Authors’ own work

Table 4.

Multivariate DCC-GARCH output table

Model component	Parameter	Coefficient	Std. err.	t-value
Joint	Dcca1	0.009	0.007	1,33
Joint	Dccb1	0.688***	0.065	10.6
No. observations	1,442
Log-likelihood	54,138.92
Av. Log-likelihood	37.54

Notes:

Multivariate DCC-GARCH model output summarizing the dcca1 and dccb1 coefficients of digital and conventional assets considered in evaluating conditional volatility during the Terra Luna crash, represented by a sample period from January 1st through May 31st, 2022. The dcca1 coefficient suggests relatively independent dynamics among the assets over time, while the dccb1 coefficient highlights a significant and persistent relationship. The significant dccb1 coefficient underscores the model’s robustness, indicating validity for conclusions drawn from the model, which provide valuable insights for risk management and portfolio diversification. The test for normal distribution was performed and showed no abnormalities. The symbols ***, ** and * denote statistical significance of t-tests at the 1, 5 and 10% level, respectively

Source: Authors’ own work

Table 5.

Univariate GARCH output table

Asset	Parameter	Coefficient	Std. err.	t-value
S&P500r	alpha1	0.026***	0.005	4.85
S&P500r	beta1	0.964***	0.005	189.51
US10Yr	alpha1	0.005***	0.001	7.94
US10Yr	beta1	0.992***	0.000	5,621.39
GOLDr	alpha1	0.016***	0.003	5.46
GOLDr	beta1	0.979***	0.002	471.69
OILr	alpha1	0.016***	0.002	9.59
OILr	beta1	0.983***	0.001	735.39
DXYr	alpha1	0.012***	0.004	3.11
DXYr	beta1	0.985***	0.002	562.55
BTCr	alpha1	0.000***	0.000	8.75
BTCr	beta1	0.995***	0.001	1,881.88
ETHr	alpha1	0.011***	0.002	6.82
ETHr	beta1	0.972***	0.006	172.41
DEFIINDEXr	alpha1	0.015***	0.002	6.51
DEFIINDEXr	beta1	0.957***	0.007	140.37
USTr	alpha1	0.071***	0.026	2.75
USTr	beta1	0.912***	0.033	27.97
LUNAr	alpha1	0.069*	0.042	1.65
LUNAr	beta1	0.93***	0.015	61.0

Notes:

This table presents the univariate GARCH coefficients alpha and beta for all digital and conventional assets of a DCC-GARCH model for the sample period surrounding the Terra Luna crash from January 1st to May 31st 2022. The high beta coefficients across all assets indicate strong volatility persistence, meaning past market shocks significantly impact current volatility. LUNA’s notably higher alpha coefficient reflects its extreme volatility during the Terra-Luna crash, highlighting the significant influence of past shocks on its variance. These findings are crucial for risk management and portfolio diversification, underscoring the importance of accounting for heightened volatility periods in strategic planning. The symbols ***, ** and * denote statistical significance of t-tests at the 1, 5 and 10% level, respectively

Source: Authors’ own work

Notes

1. Note that despite its later rebranding as LUNC, the term LUNA is used in this study to reflect the nomenclature corresponding to the sample period.

2. The correlations described and plotted in the table and figures in this section refer to volatility correlations, not return correlations.