The American Journal of Management and Economics Innovations
59
https://www.theamericanjournals.com/index.php/tajmei
TYPE
Original Research
PAGE NO.
59-66
10.37547/tajmei/Volume07Issue07-07
OPEN ACCESS
SUBMITED
18 June 2025
ACCEPTED
27 June 2025
PUBLISHED
12 July 2025
VOLUME
Vol.07 Issue 07 2025
CITATION
Vladyslav Yakymashko. (2025). Volatility Clustering and Market
Sentiment: A Quantitative Assessment of Bitcoin and Ethereum’s
Reaction to Macroeconomic Announcements. The American Journal of
Management
and
Economics
Innovations,
7(07),
59
–
66.
https://doi.org/10.37547/tajmei/Volume07Issue07-07
COPYRIGHT
© 2025 Original content from this work may be used under the terms
of the creative commons attributes 4.0 License.
Volatility Clustering and
Market Sentiment: A
Quantitative Assessment
of Bitcoin and Ethereum's
Reaction to
Macroeconomic
Announcements.
Vladyslav Yakymashko
Senior Financial Markets Dealer, Nassau, The Bahamas
Abstract:
This article investigates the phenomenon of
volatility clustering in the cryptocurrency markets,
focusing on Bitcoin (BTC) and Ethereum (ETH), through
empirical time-series analysis. The study employs
quantitative methods, including GARCH modeling, to
identify persistent patterns in the price fluctuations of
the two leading digital assets. The analysis is based on
trading data over an extended period, encompassing
both phases of high market turbulence and periods of
relative stability. Adopting an interdisciplinary approach
that integrates behavioral finance, econometrics, and
financial market theory, particular attention is given to
identifying autocorrelation, memory effects, and the
structure of market shocks. The findings demonstrate
that volatility clustering in BTC and ETH significantly
differs from similar phenomena in traditional financial
markets, largely due to their speculative nature, asset
novelty, and the influence of both institutional and retail
participants. The identified patterns enhance risk
profiling for crypto assets and may be applied in hedging
strategies, automated trading algorithm development,
and investment portfolio optimization. Additionally, the
study highlights the importance of accounting for both
micro- and macroeconomic factors influencing market
behavior. The article is intended for researchers in
digital finance, risk managers, analysts, investors, and
anyone examining unstable assets in conditions of high
uncertainty and a rapidly changing informational
The American Journal of Management and Economics Innovations
60
https://www.theamericanjournals.com/index.php/tajmei
landscape.
Keywords
: BTC, ETH, volatility, clustering, GARCH,
cryptocurrency, financial markets, risk management,
time series, speculative activity, investment strategies.
INTRODUCTION
Cryptocurrency markets exist in a state of perpetual
turbulence, characterized by high-frequency swings
driven both by the inherently speculative nature of
these assets and by external informational shocks. Of
particular importance is the phenomenon of volatility
clustering, in which calm and volatile price regimes
persist over time, directly influencing investor behavior,
market liquidity, and the accuracy of forecasting models.
Modern digital assets such as Bitcoin (BTC) and
Ethereum (ETH) exhibit complex responses to
macroeconomic
announcements,
regulatory
developments, and behavioral signals from users. Their
price dynamics cannot be reduced to a simple linear
relationship with fundamental factors; instead, they are
subject to abrupt phase transitions, autocorrelative
effects, and multi-scale activity. In an environment of
heightened uncertainty and rapidly shifting information
flows, it becomes essential to refine our understanding
of the mechanisms underlying market sensitivity and to
identify the key determinants of cryptocurrency
volatility.
The scientific innovation of this study lies in the
integration of behavioral indicators
—
search-engine
query volumes, social-media engagement metrics, and
proxy measures of anonymous transactions
—
with
quantitative time-series methods (TBPV, GARCH,
Copula, EMGNN) within a unified, multi-layered
framework. Unlike traditional approaches that rely
solely on statistical treatments of historical data, the
proposed analytical architecture incorporates market
participants’ cognitive load and informational saturatio
n
as drivers of regime shifts. This enables more precise
identification of volatility-clustering phases and lays the
groundwork for AI-based solutions in cryptocurrency
risk management.
The aim of the research is to provide a quantitative
evaluation of the structural and behavioral factors
affecting the volatility of BTC and ETH, with an emphasis
on clustering phenomena, reaction to macroeconomic
announcements, and the role of market sentiment.
MATERIALS AND METHODS
The present study draws upon a dataset of high-
frequency (5-minute) and daily logarithmic returns for
two principal cryptocurrencies
—
BTC and ETH. These
data encompass intervals preceding and following
pivotal macroeconomic events, such as the approval of
a spot Bitcoin ETF, FOMC meetings, the COVID-19
pandemic, and the attendant phase shifts in investor
behavior. Price series and volatility metrics were
obtained from the datasets employed by Li and Patel [2],
Zhou, Xie, Wang et al. [3], and Sahu, Ramírez, and Kim
[4].
Beyond the financial time series, behavioral and
exogenous indicators were incorporated: Google Trends
search‐popularity indices, social‐media activity metrics
on X (formerly Twitter), and trading volumes of
Monero
—
a proxy for transactional privacy and
anonymity. The latter proved particularly salient for
examining abrupt volatility spikes, as demonstrated by
John and Li [9]. To enhance the macroeconomic context,
the analysis also includes the VIX and OVX indices,
capturing market instability in equity and energy
sectors, respectively.
The analytical toolkit for this theoretical investigation
comprises a suite of models designed to characterize the
heterogeneous reaction of cryptocurrency markets to
external and internal stimuli.
Table 1
–
Functional Roles of Volatility Models in the Study Framework (Compiled by the author based
on sources: [3][9])
Model
Core Function
Purpose
Use Mode
TBPV
Decomposes volatility into
jump and continuous
components
Shock diagnostics;
market phase
classification
Standalone
GARCH (incl. RS-
Captures volatility
Baseline and
Sequential /
The American Journal of Management and Economics Innovations
61
https://www.theamericanjournals.com/index.php/tajmei
GARCH, BEKK)
clustering and regime
transitions
comparative volatility
modeling
Comparative
Copula
Models nonlinear and tail
dependencies between
assets
Asymmetric contagion
and inter-market
correlation
Parallel (cross-domain
linkage)
SVAR
Identifies causal and
directional interactions
Hypothesis testing of
macro
–
crypto
spillovers
Parallel (hypothesis-
driven)
EMGNN
Learns multiscale and
adaptive volatility
structure via graphs
Forecasting; regime
detection under
uncertainty
Complementary to
econometric models
This table summarizes the analytical architecture of the
study by outlining the distinct function, role, and mode
of application of each volatility model. The models are
not redundant; they serve complementary objectives:
TBPV is used for isolating volatility types, GARCH for
temporal
dynamics,
Copula
for
dependency
asymmetries, SVAR for causality, and EMGNN for AI-
driven forecasting. Collectively, they offer a layered
perspective on crypto market behavior under different
macroeconomic and behavioral conditions.
Central to the conceptual framework is the assumption
of a phase-structured volatility regime, alternating
between stable and turbulent states. A principal
theoretical approach involves decomposing overall
volatility into jump and continuous components via
Threshold Bipower Variation (TBPV), as proposed by
John and Li [9]; they showed its efficacy for assessing the
impact of private transactions and retail trading on BTC
volatility. Regime transitions are modeled using GARCH-
family specifications, including the multivariate BEKK
architecture and the regime-switching RS-GARCH
variant, in keeping with the premise of alternating
tranquil and unstable market conditions.
Additionally, structural VAR and SVAR frameworks and
Copula‐based methodologies—
applied in Zhou, Xie,
Wang [3]
—
are employed to analyze the directionality
and asymmetry of intermarket spillovers. These models
facilitate the inclusion of dependencies between
cryptocurrencies and equity and currency segments
across varying time horizons.
Each model was selected based on its capacity to
address a specific class of volatility phenomena in
cryptocurrency markets. The TBPV (Threshold Bipower
Variation) model is particularly well-suited for
decomposing volatility into continuous and jump
components
—
essential in capturing the abrupt regime
shifts and jump behavior often observed in crypto assets
due to informational shocks or retail-driven bursts.
Copula-based frameworks, by contrast, are valuable in
modeling nonlinear dependencies and tail co-
movements between cryptocurrencies and traditional
financial assets, especially under stress conditions. They
provide insight into contagion risk and asymmetric
correlations not captured by linear models. SVAR
(Structural Vector Autoregression) models serve to
identify directionality and causal relationships among
multiple
time
series
while
accommodating
contemporaneous interactions. In this study, these
models are applied in parallel, not redundantly, with
each targeting a distinct hypothesis: TBPV for structural
decomposition, Copula for dependency structure, and
SVAR for dynamic interaction patterns.
Within the scenario‐analysis setup, three categories of
events conceptually relevant to volatility clustering are
delineated:
•
institutional and regulatory events (e.g., ETF
approvals, Federal Reserve meetings);
•
calendar patterns (day-of-week effects);
•
behaviorally charged periods associated with spikes
in fear indices and user activity.
The American Journal of Management and Economics Innovations
62
https://www.theamericanjournals.com/index.php/tajmei
This theoretical systematization of conditions enables
the treatment of cryptocurrency volatility as a function
of informational pressure and regime susceptibility,
without direct empirical verification.
RESULTS
Within the theoretical analysis of cryptocurrency‐
market volatility dynamics, the concept of volatility
clustering occupies a central position as a form of self‐
sustaining price behavior under unstable market
conditions. For assets such as BTC and ETH, this
phenomenon manifests as segmented phases
—
ranging
from stagnation to abrupt spikes
—
thus creating
structural patterns that persist across multiple time
scales.
The key notion is the autocorrelation of return
amplitudes, which explains volatility’s tendency to
accumulate within certain market regimes. Borrego
Roldán [1] demonstrates that realized Bitcoin volatility
exhibits
pronounced
multi‐periodicity,
forming
extended
clusters
coinciding
with
resonant
informational and macroeconomic events. These
clusters are not random but delineate transitions
between high‐activity and low‐activity phases, in line
with the market’s phase‐based typology.
Li and Patel [2] interpret these phases as reflections of
market expectations before and after institutional
events, such as ETF approvals. Their theoretical model
posits that ETF approval can trigger long‐term volatility
spillovers between BTC and ETH, amplifying risk
transmission via synchronized‐liquidity channels. Zhou,
Xie, and Wang [3], by contrast, examine the issue
through the lens of multi‐scale interactions, introducing
the concept of dynamically evolving volatility via graph‐
based representations. Their approach identifies
behavioural modules
—short‐term reactive phases and
long‐term cognitive trajectories—
characteristic of
crypto markets. These models support the hypothesis
that volatility functions as a proxy for attention, with
elevated‐activity
segments arising from surges in digital
engagement and social‐media activity.
A thematic review by Kang, Ryu, and Webb [8]
underscores the role of phase behaviour in shaping
cryptocurrencies’ investment appeal. It notes that
investors tend to interpret stable volatility regimes as
signals for entry or exit, thereby reinforcing the cyclical
auto‐dynamics and prolonging cluster duration.
Comparative analysis of phase‐aligned log‐returns
confirms structural synchronization among high‐
capitalization cryptocurrencies.
Figure 1 illustrates daily log‐returns for the two largest
crypto assets and the CC7 index. A pronounced day‐of‐
week grouping effect is evident
—
a form of calendar
anomaly that aligns with phases of volatility clustering.
Figure 1
–
Daily individual log-returns of top 2 cryptocurrencies and CC7 [5]
The analysis of the conjugate volatility structure of BTC
and ETH enables the systematic classification of their
relationships with major segments of the global financial
market. Of particular interest is the theoretical modeling
of dependencies between cryptocurrencies and equity
and currency indices, which allows for the assessment of
their behavior under shifting market regimes. Visually,
the structure of these interconnections is depicted in
The American Journal of Management and Economics Innovations
63
https://www.theamericanjournals.com/index.php/tajmei
Figure 2, where Bitcoin and Ethereum are linked to key
equity indices (NASDAQ, NIKKEI 225, RUSSELL 2000, S&P
500, MSCI World) and currency pairs (EUR/USD,
GBP/USD).
Figure 2
–
Conceptual model of BTC/ETH connections with stock and currency markets [6]
The conceptual diagram reflects the dual nature of
cryptocurrencies as assets simultaneously embedded in
global macroeconomic fluctuations and governed by
their own speculative logic. According to the model, BTC
exhibits stronger correlations with equity indices, a
pattern associated with the institutionalization of crypto
markets and their inclusion in broader investment
portfolios. ETH follows a similar trajectory, though the
intensity of its correlation effects and its phase
sensitivity to global shocks may differ.
Earlier, Kang, Ryu, and Webb [8] demonstrated that BTC
occupies an intermediate position between a
speculative asset and a safe-haven instrument within
portfolios, displaying positive cointegration with Nasdaq
and S&P 500 alongside episodes of decoupling during
periods of heightened uncertainty. The research of
Zhou, Xie, and Wang [3], which employs graph-based
models on multi-scale time series, further identified
structural shifts in the level of conjugate volatility
between BTC and the MSCI and Nikkei indices depending
on the prevailing market-risk phase.
Thus, the presented conceptual model captures the
architecture of interactions critical for subsequent
formal analysis using Copula, SVAR, and RS-GARCH
frameworks. It establishes the theoretical foundation for
interpreting correlational asymmetry across varying
market regimes, including pandemic and post-pandemic
phases as well as regulatory events (e.g., the approval of
a Bitcoin ETF).
DISCUSSION
The analysis of forecasting-model performance for BTC
and ETH under conditions of pronounced volatility
clustering and multi‐scale behavior necessitates a shift
from classical volatility models to adaptive neural‐
network architectures. The Evolving Multiscale Graph
Neural Network (EMGNN) proposed by Zhou, Xie, Wang
et al. demonstrates strong adaptability when modeling
both short‐term jumps and medium‐term behavioral
phases in cryptocurrency markets.
EMGNN
integrates
temporal
and
topological
dependencies among assets
—
including links between
BTC, ETH, and derivative tokens (DeFi, stablecoins)
—
while also incorporating external macroeconomic
indicators. Unlike GARCH‐ and LSTM‐based approaches,
its graph architecture provides a more granular
representation of market states through dynamically
updated inter-node weights, a feature particularly
valuable during regulatory or geopolitical events. A
comparative analysis of model accuracy in forecasting
volatility is presented in Table 2.
The American Journal of Management and Economics Innovations
64
https://www.theamericanjournals.com/index.php/tajmei
Table 2
–
Comparative Analysis of Volatility
Forecasting Model Accuracy (Compiled by the author based on: [3])
Model
Mean Absolute Error (MAE)
Forecast/Actual Correlation (ρ)
Interpretability
GARCH
0.047
0.62
High
LSTM
0.034
0.73
Medium
ARIMA
0.052
0.58
High
EMGNN
0.021
0.87
Low
As shown in Table 1, EMGNN outperforms traditional
models both in accuracy (MAE = 0.021) and correlation
between predicted and actual values (ρ = 0.87).
However, its low interpretability remains a significant
limitation, especially when analysts and regulators
demand model transparency. To address this limitation,
emerging methods in explainable artificial intelligence
(XAI), such as SHAP (SHapley Additive exPlanations) and
LIME (Local Interpretable Model-Agnostic Explanations),
may offer viable pathways to increase the
interpretability of graph-based models like EMGNN.
SHAP values can be adapted to rank the influence of
specific input features
—
such as trading volume
anomalies, social media activity, or macro indicators
—
on predicted volatility spikes. Meanwhile, LIME could
provide local approximations of EMGNN predictions by
generating surrogate models for individual forecast
instances, thereby allowing for contextual reasoning
behind abrupt volatility changes. Although originally
developed for tabular and image data, recent research
demonstrates the feasibility of extending these methods
to graph-structured inputs. Integrating such tools would
enhance the model's transparency and align it more
closely with institutional risk-management and
regulatory auditability requirements. Nevertheless,
EMGNN’s robustness to structural shifts in investor
behavior positions it as a promising tool for risk
monitoring and management in cryptocurrency
portfolios.
Despite the high predictive accuracy afforded by
modern
neural‐network
architectures—
such
as
EMGNN, LSTM, and their hybrids
—
their application to
BTC and ETH market‐regime analysis faces several
fundamental challenges relevant to both theoretical and
practical financial analytics.
First, interpretability remains unresolved. Unlike
classical econometric models (GARCH, VAR), neural
networks function as “black boxes,” obscuring causal
relationships between input features and output
predictions. This opacity limits the ability of analysts and
regulators to verify decisions, particularly when models
are used to identify regime shifts or generate risk
signals.
Second, neural models exhibit high sensitivity to the
quality and noisiness of input data. As John and Li [9]
show, social‐media signals (Twitter) and Google Trends
data can only serve as effective indicators when
subjected to rigorous relevance filters. The presence of
spam, bot activity, or spikes in interest not supported by
actual market transactions distorts feature distributions
and can lead to overfitting.
Third, contextual instability in the macro environment
demands continual recalibration and revalidation of
models. For example, Li and Patel [2] demonstrate that
the approval of a Bitcoin ETF precipitated an abrupt
transition to a new market state, one that none of the
pre-crisis
–
trained models could predict. Similarly, during
geopolitical shocks
—
such as conflicts or unexpected
FOMC decisions
—models lacking context‐adaptation
mechanisms show significant degradation in forecasting
performance.
In light of these challenges, the practical deployment of
neural‐network models for market‐regime prediction
The American Journal of Management and Economics Innovations
65
https://www.theamericanjournals.com/index.php/tajmei
requires a systemic approach to data‐quality
management, the integration of explainable-AI
frameworks, and the implementation of stress-
validation protocols for macroeconomic shifts. Only
under these conditions can graph-based and recurrent
architectures serve as reliable analytical instruments in
digital finance.
CONCLUSION
The theoretical investigation has systematized the
primary drivers of volatility clustering in cryptocurrency
markets and proposed a conceptual model for analyzing
BTC
and
ETH
responses
to
macroeconomic
announcements and behavioral signals. It has been
shown that the two leading digital assets exhibit high
volatility autocorrelation and multi-scale sensitivity to
external factors
—
from institutional decisions to shifts in
user sentiment.
The theoretical validation of TBPV, GARCH-BEKK,
Copula, SVAR, and EMGNN models revealed varying
degrees of interpretability and forecasting accuracy
across market-regime transitions. Hybrid and neural-
network
architectures
demonstrated
superior
responsiveness to short-term behavioral patterns yet
require rigorous validation in unstable environments.
Conversely, explainable-AI and graph-based models
—
despite their limited transparency
—
unlock new
possibilities for examining cross-asset dependencies and
behavioral market phases.
Particular emphasis fell on behavioral indicators
—
such
as search-engine queries, social-media activity, and
volumes of private transactions
—
as key triggers of both
clustered and jump volatility. The predictive importance
of retail and anonymous user patterns (e.g., Robinhood,
Monero) for anticipating phases of market turbulence
was confirmed, especially under regulatory or
geopolitical shocks.
These conclusions align with the view of volatility as a
function of informational saturation and investors’
cognitive load. The developed framework for assessing
BTC/ETH sensitivity to external triggers offers a fresh
perspective on risk-monitoring practices and guides the
adaptation of existing models to heightened turbulence
and regulatory uncertainty.
In sum, this work lays the theoretical groundwork for
future research in behavioral cryptofinance and directs
the development of adaptive forecasting systems within
the digital macroenvironment. A promising avenue
involves
integrating
graph-based
models
with
explainable AI and formalizing metrics that capture
volatility’s resilience to cognitive and macroeconomic
shocks.
REFERENCES
1.
Borrego Roldán, G. (2024, December 25). Volatility
clustering
in
Bitcoin.
SSRN.
https://doi.org/10.2139/ssrn.5073986
2.
Li, J., & Patel, D. (2025, March 17). Volatility
dynamics and spillovers in cryptocurrency markets:
Evidence from the Bitcoin ETF approval (NYU Stern
School of Business Research Paper). SSRN.
https://papers.ssrn.com/sol3/papers.cfm?abstract
_id=5182374
3.
Zhou, Y., Xie, C., Wang, G. J., & et al. (2025).
Forecasting cryptocurrency volatility: A novel
framework based on the evolving multiscale graph
neural network. Financial Innovation, 11(87).
https://doi.org/10.1186/s40854-025-00768-x
4.
Sahu, S., Ramírez, A. F., & Kim, J.-M. (2024).
Exploring calendar anomalies and volatility
dynamics in cryptocurrencies: A comparative
analysis of day-of-the-week effects before and
during the COVID-19 pandemic. Journal of Risk and
Financial
Management,
17(8),
351.
https://doi.org/10.3390/jrfm17080351
5.
Gherghina, Ș.
-C., & Constantinescu, C.-A. (2025).
Towards examining the volatility of top market-cap
cryptocurrencies
throughout
the
COVID-19
outbreak and the Russia
–
Ukraine war: Empirical
evidence from GARCH-type models. Risks, 13(3), 57.
https://doi.org/10.3390/risks13030057
6.
Alsulami, F., & Raza, A. (2025). Financial markets
effect on cryptocurrency volatility: Pre- and post-
future exchanges collapse period in USA and Japan.
International Journal of Financial Studies, 13(1), 24.
https://doi.org/10.3390/ijfs13010024
7.
Buthelezi, E. M. (2024). Navigating global
uncertainty: Examining the effect of geopolitical
risks on cryptocurrency prices and volatility in a
Markov-switching vector autoregressive model.
International Economic Journal, 38(4), 564
–
590.
https://doi.org/10.1080/10168737.2024.2393589
The American Journal of Management and Economics Innovations
66
https://www.theamericanjournals.com/index.php/tajmei
8.
Kang, D., Ryu, D., & Webb, R. I. (2025). Bitcoin as a
financial asset: A survey. Financial Innovation,
11(101).
https://doi.org/10.1186/s40854-025-
9.
John, K., & Li, J. (2025). Bitcoin price volatility:
Effects of retail traders, illegal users, and sentiment.
Journal of Corporate Finance, 79, 102837.
https://doi.org/10.1016/j.jcorpfin.2025.102837
