Analyzing Long Memory in Cryptocurrency Volatility: Bitcoin, Ethereum, and Ripple

Introduction

The global financial landscape has undergone significant transformation since the early 2000s. A growing loss of confidence in traditional, central bank-based fiat currency systems has emerged, driven by factors such as loose monetary policies, global imbalances, and the resulting financial crises. These events increased market volatility and shifted investor risk perceptions, creating an environment ripe for alternative financial instruments.

Simultaneously, advancements in blockchain technology paved the way for cryptocurrencies. Bitcoin, the first and most prominent cryptocurrency, garnered substantial attention for its fully digital nature, fast and low-cost transaction systems, and decentralized control. Its rise was followed by numerous other cryptocurrencies, with Bitcoin, Ethereum, and Ripple collectively dominating a significant portion of the market capitalization and trading volume.

A critical characteristic of these digital assets is their exceptionally high volatility, which surpasses that of traditional assets like stocks or major fiat currencies. This volatility makes accurate modeling crucial for risk management and investment decisions. This study examines the long memory properties in the volatility of these three major cryptocurrencies, supporting the use of specialized modeling techniques for better risk assessment.

Understanding Cryptocurrency Volatility and Long Memory

Volatility represents the degree of variation in an asset's trading price over time. For investors and portfolio managers, it is a fundamental measure of risk. Traditional methods, like standard deviation, often fall short in capturing the complex, time-varying nature of volatility in modern financial markets.

Generalized Auto Regressive Conditional Heteroscedasticity (GARCH) models have become a standard tool for estimating volatility. These models account for time-varying volatility by incorporating historical conditional variance and market shocks. However, the unique characteristics of cryptocurrency returns—such as high excess kurtosis (fat tails) and negative skewness—require more specialized approaches.

Long memory, or long-range dependence, refers to a statistical phenomenon where observations far apart in time remain correlated. In financial terms, it means that volatility shocks persist for long periods, decaying at a hyperbolic rate rather than quickly. This property contradicts the efficient market hypothesis, which assumes price changes are random and unpredictable. The presence of long memory suggests that future volatility can be predicted to some extent, offering potential advantages for forecasting and risk management.

👉 Explore advanced volatility forecasting tools

Methodology: Testing for Long Memory

To investigate long memory in cryptocurrency volatility, this study employed four established statistical tests on daily return data for Bitcoin, Ethereum, and Ripple:

• Rescaled Range (R/S) Statistics: Developed by Hurst and later modified by Lo, this test measures the scaling behavior of the range of cumulative deviations from a time series' mean. Lo's modified version adjusts for potential short-term dependence.
• Geweke and Porter-Hudak (GPH) Model: A semi-nonparametric approach that uses spectral analysis to estimate the fractional integration parameter (d), which indicates the presence and degree of long memory.
• Gaussian Semiparametric (GSP) Method: This technique, developed by Robinson and Henry, specifies the shape of the time series' spectral density to estimate the long memory parameter.

The tests were applied to both the daily returns and the squared daily returns (a common proxy for daily volatility) of each cryptocurrency.

Key Findings on Long Memory

The analysis yielded clear and significant results:

• Daily Returns: The null hypothesis of no long-range dependence could not be rejected for the daily return series of Bitcoin, Ethereum, and Ripple. This indicates that the returns themselves do not exhibit a strong long memory effect.
• Squared Returns (Volatility): In contrast, the tests provided strong evidence of long memory in the squared returns for all three cryptocurrencies. The estimated fractional integration parameter (d) was statistically significant and fell within the range of 0 to 0.5, confirming a long memory process in their volatility.

This finding is consistent with previous research focused solely on Bitcoin and underscores that volatility clustering and persistence are fundamental features of major cryptocurrency markets.

Optimal GARCH Models for Each Cryptocurrency

Given the confirmed presence of long memory in volatility, standard GARCH models are insufficient. This study estimated several fractional GARCH extensions to identify the best fit for each digital asset:

• Fractional Integrated GARCH (FIGARCH): This model captures the slow, hyperbolic rate of decay in volatility shocks, recognizing both long and short memory characteristics.
• Hyperbolic GARCH (HYGARCH): An extension of FIGARCH that introduces weights into the difference operator, allowing for modeling long memory with hyperbolic convergence rates and permitting the existence of second moments.

The models were evaluated using student-t and skewed student-t distributions to account for the fat-tailed nature of cryptocurrency returns. The best-fitting models were selected based on information criteria like Akaike (AIC), Schwarz (SW), and Hannan-Quinn (H-Quinn), along with log-likelihood statistics.

• Bitcoin (BTC): The HYGARCH model with a student-t distribution provided the smallest values for all information criteria, indicating it was the best fit for modeling Bitcoin's volatility.
• Ethereum (ETH): The FIGARCH model with a skewed student-t distribution produced the best estimations. A positive and significant asymmetry parameter indicated that Ethereum returns are skewed to the right.
• Ripple (XRP): The FIGARCH model with a student-t distribution was identified as the most suitable model for Ripple returns.

Value at Risk (VaR) and Expected Shortfall Analysis

To assess the practical risk management application of these models, the study conducted Value at Risk (VaR) backtesting. VaR estimates the potential loss in value of a portfolio over a defined period for a given confidence interval. The accuracy of the VaR models was tested using Kupiec's Proportion of Failures (POF) test.

• Backtesting Results: The FIGARCH and HYGARCH models performed well for most quantiles and for both long and short trading positions. The Kupiec test statistics were not significant in most cases, indicating that the empirical failure rates aligned with the pre-specified VaR levels. This confirms that these long-memory models are effective for predicting risk in cryptocurrency markets.

Furthermore, the study calculated Expected Shortfall (ES), a coherent risk measure that predicts the expected loss conditional on the loss being larger than the VaR. The results showed:

• Expected shortfalls were highest for Ripple and lowest for Bitcoin across all risk levels.
• This implies that investors may require more capital to cover potential losses when dealing with Ethereum and Ripple compared to Bitcoin.

👉 Learn more about managing crypto investment risk

Conclusion and Implications for Investors

This study confirms the presence of significant long memory in the volatility of Bitcoin, Ethereum, and Ripple. This finding has crucial implications:

1. Market Efficiency: The presence of long memory implies a degree of market inefficiency, meaning future volatility is somewhat predictable. This allows investors to potentially capture speculative profits and improve risk-adjusted portfolio performance.
2. Model Selection: Standard volatility models are inadequate. Fractional GARCH extensions like FIGARCH and HYGARCH, which account for long memory, asymmetry, and fat tails, provide superior forecasting accuracy for cryptocurrency risk.
3. Risk Management: The successful VaR backtesting demonstrates that these models are reliable tools for financial institutions and investors to measure and manage the heightened risk associated with cryptocurrency investments.
4. Diversification: Despite their high volatility, cryptocurrencies' low correlation with traditional assets still makes them a compelling option for portfolio diversification. However, investors must be aware of the significant capital requirements needed to cover potential losses, especially with assets like Ethereum and Ripple.

As cryptocurrencies continue to mature and potentially become mainstream financial instruments, understanding and accurately modeling their unique volatility dynamics will be paramount for the future of digital asset investment and risk management.

Frequently Asked Questions (FAQ)

What is long memory in volatility?

Long memory, or long-range dependence, is a property where volatility shocks have a persistent effect that decays very slowly over time (hyperbolically). This means past volatility can influence future volatility for a long period, making it partially predictable.

Why are GARCH models used for cryptocurrency volatility?

Cryptocurrency returns exhibit time-varying volatility, clustering (high volatility tends to be followed by more high volatility), and fat tails. GARCH models are specifically designed to capture these characteristics, making them more suitable than simpler measures like standard deviation.

What is the difference between FIGARCH and HYGARCH?

FIGARCH (Fractional Integrated GARCH) directly models the long memory component. HYGARCH (Hyperbolic GARCH) is an extension that provides more flexibility by adding weights to the model, often leading to better performance, especially for assets with extreme volatility like Bitcoin.

How is Value at Risk (VaR) used in this context?

VaR is a risk measure that estimates how much a cryptocurrency investment might lose, with a given probability, over a specific time frame. This study backtested VaR estimates from the long-memory models to prove their accuracy in predicting potential losses.

What does a higher Expected Shortfall mean for an investor?

Expected Shortfall calculates the average loss that could occur on the worst days, beyond the VaR threshold. A higher Expected Shortfall, as seen with Ripple and Ethereum, indicates a higher potential for extreme losses, meaning an investor would need to set aside more capital to cover this risk.

Can these findings be applied to other cryptocurrencies?

While this study focused on the three largest cryptocurrencies, the methodology and the general finding of long memory in volatility likely apply to other major cryptocurrencies with similar trading volumes and market structures. However, the specific optimal model (e.g., FIGARCH vs. HYGARCH) may differ.