TL;DR
This paper introduces a sophisticated hybrid framework—combining eGARCH, Extreme Value Theory (EVT), and Copula functions—to analyze risk spillovers between cryptocurrencies and global financial indices. The study reveals that while cryptocurrencies are highly interconnected with one another, they remain surprisingly decoupled from traditional global markets, offering significant diversification benefits but limited hedging potential for investors.
What Problem Does This Paper Address?
As cryptocurrencies become a staple in modern portfolios, understanding their "spillover" effects—how volatility in one market infects another—is critical for risk management. Traditional risk measures like Value at Risk (VaR) and Expected Shortfall (ES) have well-known flaws: VaR ignores the severity of extreme losses, while ES lacks "elicitability" (making it hard to compare model performance) and robustness against outliers.
The authors tackle two main challenges:
- Modeling "Fat Tails": Financial returns, especially in crypto, don't follow a normal distribution; they have extreme peaks and "fat tails" (frequent extreme events).
- Regulatory Arbitrage: When different risk models yield vastly different capital requirement estimates for the same portfolio, it creates "regulatory arbitrage." The paper explores Range Value at Risk (RVaR) as a more robust, elicitable alternative to ES to mitigate this.
Methods
The researchers employ a multi-stage hybrid approach that improves upon standard "black-box" correlations:
- Volatility Modeling: They compare symmetric (sGARCH) and asymmetric (eGARCH, gjrGARCH) models. The eGARCH model proved best at capturing how "bad news" impacts volatility differently than "good news."
- Bulk-and-Tails (BATs) Model: Instead of standard Extreme Value Theory (EVT) which requires choosing an arbitrary "threshold" for extreme events, they use the novel BATs parametric model. This flexible distribution models the entire return series—both the "bulk" (normal returns) and the "tails" (extreme events)—simultaneously.
- Copula Functions: To link individual asset risks into a joint multivariate distribution, they test several Copula families (Frank, Gumbel, Joe, and Student’s $t$). This allows them to model complex, non-linear dependencies between assets.
- Spillover Analysis: They use the Diebold-Yilmaz (2012) approach, based on forecast error variance decomposition, to quantify exactly how much "risk" is exported from one asset to another.
Datasets
The study uses daily closing prices from January 1, 2018, to March 31, 2024, sourced from Yahoo Finance. The assets include:
- Cryptocurrencies: Bitcoin (BTC) and Litecoin (LTC).
- Global Indices: MSCI World Index (representing developed markets) and MSCI Emerging Markets Index.
Key Insights and Ideas
- The "Crypto Island": The most striking finding is the lack of significant spillover between crypto and traditional indices. Cryptocurrencies are "net exporters" of risk to each other but have a minimal impact on the broader global financial system.
- Diversification vs. Hedging: Because cryptos move relatively independently of global indices, they are excellent for diversification (reducing overall portfolio variance). However, they are poor hedges; they don't necessarily move in the opposite direction of traditional stocks during a crash—they just do their own thing.
- RVaR as a Regulatory Hero: The results highlight that RVaR is superior to ES in terms of "legal robustness." It is less sensitive to model misspecification, meaning different banks using different internal models would arrive at more consistent capital requirements using RVaR than ES.
- Tail Flexibility: The use of the BATs model confirms that crypto returns have highly asymmetric tails, and modeling them without fixed thresholds provides a much more accurate picture of "black swan" risks.
Key Results
- Model Accuracy: The eGARCH-EVT-Copula framework outperformed conventional methods (like Historical Simulation) for VaR estimation.
- Spillover Metrics: The total spillover index for the four-asset system was 45.67%.
- Specific Interdependence: Bitcoin’s spillover to the MSCI World Index was a low 6.68%, whereas its spillover to Litecoin was a much higher 33.55%.
- Loss Function Performance: While the hybrid model won for VaR, Historical Simulation (HS) remained surprisingly competitive for ES, and the $t$-distributed parametric method performed best for RVaR in terms of realized loss values.
Limitations and Future Work
- Asset Variety: The study only looked at Bitcoin and Litecoin. The authors suggest that future research should include a wider array of digital assets, such as Ethereum, DeFi tokens, or stablecoins.
- Dynamic Dependencies: The current copula approach could be expanded to "time-varying" copulas to see how the relationship between crypto and stocks shifts during specific geopolitical crises or "crypto winters."
- Broader Asset Classes: The authors recommend testing this framework against other traditional hedges like gold, oil, or treasury bills to see if the "diversification benefit" holds up across all alternative investments.
Summary updated