Marjan Sadat Fatemi Ghomi, Abbas Saghaei, Majid Mirzaee Ghazani
Accurately forecasting cryptocurrency price movements following market overreactions is crucial for traders, investors, and risk managers operating in highly volatile environments. This study presents a novel multi-source, multi-output deep learning framework designed to predict the direction of price changes in four major cryptocurrencies — Bitcoin (BTC), Ethereum (ETH), Binance Coin (BNB), and Ripple (XRP) — immediately after overreaction events. By jointly modeling these assets, our approach captures their interconnected market dynamics, enhancing predictive accuracy. We compile an extensive dataset with over 656 features from diverse sources, including historical trading data, on-chain metrics, technical indicators, and social sentiment data from Google Trends, collected at both daily and intraday frequencies. To improve model interpretability and performance, we introduce two engineered features — price change magnitude and price variation speed — that effectively represent intraday volatility. Feature selection using a Random Forest approach reduces the feature set to 30 key variables, ensuring robustness and avoiding overfitting. Using three advanced deep learning architectures — LSTM, RNN, and CNN — we train models to classify the next-day price movement as upward or downward. Empirical results demonstrate that the multi-output LSTM achieves an F1-score of 73.42%, outperforming both single-asset models (62.95–68.25%) and alternative architectures. These findings highlight the benefits of joint modeling, leading to more reliable forecasts during turbulent market conditions. Our framework offers a practical tool for algorithmic trading, portfolio management, and risk mitigation in the dynamic cryptocurrency landscape.
The paper provides an integrated literature review of recent scientific publications on quantum computing in finance and identifies promising directions for future research on the subject. The review covers seven thematic areas: portfolio optimization, derivative pricing and stochastic volatility, quantum machine learning for fraud detection and credit risk, insurance and actuarial science, mixed-frequency econometrics, fuzzy-quantum approaches for financial explainability, and security of cryptocurrencies. The paper compiles the essential quantum computational methods proposed in the literature, outlines their economic significance and the existing constraints for empirical testing and implementation, and discusses cross-cutting issues of explainability, trustworthy AI, robustness, and governance that arise across these application domains. Drawing on this review, the paper identifies five macro-gaps in the existing literature and proposes seven concrete directions for future research, grounded in European financial data and currently available quantum computing infrastructure. A special focus throughout is the increasingly available quantum infrastructure in Europe and the regulatory emphasis on trustworthy artificial intelligence, both of which create timely opportunities for future applications in financial modelling, risk management, and explainable financial AI.
Abstract This study examines the dynamic volatility spillover between financial technologies and traditional and alternative financial markets. The analysis utilizes daily data covering the period from January 2018 to March 2026 for FinTech ETFs, the Nasdaq, Bitcoin, and gold markets. Interconnectivity among financial markets was analyzed using the time-varying parameter VAR (TVP-VAR) connectedness approach. The findings indicate the presence of a moderate and time-varying connectivity structure among the markets. In particular, a strong interaction was observed between the FinTech and Nasdaq markets, while Bitcoin was found to play a significant role as a volatility transmitter during certain periods. The gold market, on the other hand, generally exhibited more stable and limited interactions. Additionally, the study found that financial market linkages increase during periods of crisis and uncertainty. These results highlight that the transmission of volatility across financial markets has a dynamic structure and that portfolio diversification strategies should be evaluated accordingly. By examining FinTech markets alongside other major asset classes, the study provides a timely and comprehensive contribution to the literature.
Bitcoin has become an increasingly important asset for portfolio allocation, yet its diversification value and option-implied information remain difficult to evaluate. This paper examines Bitcoin risk from portfolio and option-implied perspectives. This study assesses whether Bitcoin improves the risk-return opportunity set with traditional assets and whether its diversification role remains stable during market stress. Option-implied measures, including the 25-delta Risk Reversal (RR25), smile curvature, and an at-the-money Implied-Volatility-minus-Realized-Volatility (IV-minus-RV) proxy, are then constructed to predict market conditions. Portfolio analysis shows that Bitcoin can improve risk-return tradeoffs but does not function as a stable minimum-variance asset or a reliable crisis hedge. Baseline regressions provide limited evidence that RR25 consistently predicts future realized volatility or returns. However, extreme negative short-dated RR25 is followed by higher future realized volatility, suggesting RR25 is more informative as a nonlinear stress-state indicator than as a continuous forecasting variable. Smile curvature captures the implied-volatility surface but provides weaker predictive information. Finally, the IV-minus-RV analysis shows that gradual RR25-based exposure scaling achieves a better risk-adjusted profile than a binary exposure rule. Overall, the findings indicate that Bitcoin's diversification benefits and option-implied information are state-dependent.
A Universal, Options-Free, One-Byte Market-State Primitive: Cross-Asset Transfer, Distributional-Novelty Detection, and Privacy-Preserving Federation Randolph James Ferlic, M.D. and Kimberly Kate Ferlic (Fieldstone Analytics, LLC, Austin, TX, USA) Preprint · Zenodo DOI: 10.5281/zenodo.22116173 · CC-BY 4.0 · Community: spiral-domain-encoder-campaign Abstract A companion study established that a fixed class-discriminant encoder — which reduces a window of a multivariate stream to a single 8-bit token — is a near-lossless detector of market stress across asset classes, but a taxed classifier and a null forecaster. This study asks, under strict pre-registration, whether that detector becomes a *universal, deployable primitive*, and finds that it does, along five axes, with an honest boundary preserved throughout. (1) The codebook is universal: fit on equities alone, it detects stress zero-shot across cryptocurrency, commodities, emerging-market funds, credit, and foreign exchange at a median AUC of 0.84, within 0.01 of a bespoke per-asset codebook, because the calm-to-stress discriminant direction is nearly identical across asset classes (median cross-asset cosine 0.94). (2) The token detects distributional regime-change that volatility structurally misses — the one place it beats volatility — and is an orthogonal channel that improves a volatility-only stack by +0.18 AUC on a mixed target. (3) It generalizes to credit and rates (pooled AUC 0.91) and to currency crosses (0.80), including markets with no options index. (4) It is cheap at scale and low-data: a genuine one-byte-on-the-wire firm-wide dashboard detects systemic stress at 0.92 at 200× compression — a 42-instrument universe fits in ~10 kilobytes per year — and a new deployment is usable after ~3 months of calm history. (5) The same one-byte token supports privacy-preserving federation: institutions each sharing one token per day match full-data systemic-detection utility while their individual positions are unrecoverable even to a non-linear, colluding, market-informed adversary (incremental reconstruction R² ≈ 0.01). All headline results survive a permutation-placebo leakage test, moving-block bootstrap confidence intervals, and robustness to split, seed, stress-label, and a strict recent holdout. Every negative is reported, and two honest scope refinements are disclosed: a formal differential-privacy guarantee is event-level, not stream-level, and portfolio membership is coarsely inferable while positions are not. Highlights · Universal codebook — one equity-fit codebook detects stress zero-shot across every asset class (median 0.84, transfer loss ~0.01, 22/23 targets); mechanism = a shared cross-asset stress direction (cosine 0.94). · A capability volatility lacks — the token detects distributional (shape) regime-change where trailing volatility is at or below chance; an orthogonal channel that complements any GARCH/VIX stack. · Coverage — credit/rates (pooled 0.91) and FX/carry white-space (0.80), options-free, at one byte per instrument. · Compression-at-scale — a 1-byte-on-the-wire firm-wide dashboard at AUC 0.92, 200× compression, ~10 KB/instrument-year; detection-sufficient encoder ≈ 590 multiply-accumulates and < 3 KB (microcontroller-class). · Low-data cold-start — usable after ~3 months of calm data; saturates within ~2 years; train on a trailing window. · Privacy-preserving federation — full-data-equivalent systemic detection with individual positions non-invertible even under a strong adversary; optional event-level differential privacy; manipulation-robust median aggregation; per-channel attribution; a yield-curve-shape token. · Information-theoretic characterization — detection is a ~1-bit decision while classification is high-bit; competitive with principled two-sample/OOD tests at O(1) online cost. · Honest negatives and hardening — network topology, temporal-token, cross-sectional ranking, and tail-edge results are all reported null; every headline result is leakage-tested (permutation placebo → chance) and confidence-bounded. What this record contains · Manuscript_Paper40.pdf — the manuscript (single-column preprint). · PAPER_40_ZENODO_ARCHIVE.zip — the reproducibility archive: seven pre-registration scopes (frozen outcome bands) covering studies FIN-15 … FIN-45; the deterministic per-study runners and the shared frozen encoder/loader and federation modules; the per-study JSON result summaries behind every figure and table value; the figure-rebuild script; and the eight figures. All paths and identifiers are scrubbed and leak-scanned per the campaign deposit discipline; no raw market data is redistributed. Cite as R. J. Ferlic and K. K. Ferlic, "A universal, options-free, one-byte market-state primitive: cross-asset transfer, distributional-novelty detection, and privacy-preserving federation," Zenodo, 2026, doi: 10.5281/zenodo.22116173. License and patent notice Released under the Creative Commons Attribution 4.0 International License (CC-BY 4.0). Consistent with that license, no patent, patent-application, or other intellectual-property right of the authors is licensed, waived, granted, or otherwise conveyed by this deposit; the methods described — including the single-token class-discriminant encoder, its multi-token product-quantization variant, its unsupervised nearest-centroid-distance monitoring mode, and the privacy-preserving multi-party aggregation of its tokens — are the subject of filed and pending U.S. patent applications. Licensing inquiries: randolphf@fieldstoneanalyticsllc.com. Companion deposits (spiral-domain-encoder-campaign) · Single-token financial market-state monitor (companion study): doi:10.5281/zenodo.22101085 · Single-token industrial sensor substrate: doi:10.5281/zenodo.20854722 · Hardening and generality characterization: doi:10.5281/zenodo.20802759 · Deterministic multi-token token ladder: doi:10.5281/zenodo.22003179 Keywords universal codebook; decision-preserving compression; distributional novelty; federated privacy; differential privacy; systemic risk; market-stress detection; class-discriminant codebook; cross-asset transfer; edge computing; anomaly detection; pre-registration; honest negatives; financial time series; regime detection; credit risk; yield curve; volatility regime; VIX; market surveillance
Serhat Yüksel, Gabriela Oana Olaru, Serkan Eti, Hasan DİNÇER
It is frequently emphasized in the behavioral finance literature that investment decisions cannot be explained solely by economic indicators and rational expectations and that psychological factors also play a significant role in this process. However, the lack of a comparative analysis of the importance of psychological factors influencing investor behavior in the literature and the lack of consensus on which factors are more dominant constitute a fundamental problem. This deficiency leads to significant uncertainties in both theoretical modeling and practical investment strategies, increasing market risks such as irrational price movements, speculative bubbles, and panic selling. In this context, the aim of this study is to determine the relative importance of the fundamental psychological factors influencing investor decisions and, considering these factors, to identify the most appropriate investment alternatives for individuals. This study develops a new integrated decision-making model to answer these research questions. Considering the demographic characteristics of the experts, importance coefficients are calculated using the Euclidean distance-based weighting approach. Criterion weights are then determined using the Entropy method, and the MABAC and MAIRCA methods are applied to rank investment alternatives. Additionally, fractal fuzzy sets based on the Sierpinski triangle are integrated into the proposed model to model uncertainty more effectively. The study's contributions to the literature are highlighted in three dimensions: (1) psychological factors, often overlooked in the literature, are included in the criteria set; (2) expert weights are differentiated based on demographic characteristics rather than assumed to be equal; and (3) expert opinions are modeled more flexibly and precisely using new fractal number-based fuzzy sets. The findings indicate that trust is the most critical psychological factor, followed by loss aversion. In terms of investment alternatives, stocks stand out as the most suitable option, while bonds/deposits and gold are other important alternatives, with cryptocurrencies and real estate ranking next.
Liu Jin, Yahya M.H., Saidatunur Fauzi Saidin, Li Lu
Abstract Multivariate cryptocurrency forecasting is challenging because market series exhibit non-stationarity, cross-variable dependence, heterogeneous temporal scales, and abrupt short-term fluctuations. Although Transformer-based forecasting models can capture long-range temporal relationships, directly modeling raw high-frequency sequences may obscure dominant periodic structures and increase computational cost. This study proposes a frequency-guided multi-scale decomposition and patch Transformer, termed FMDP-Transformer, for multivariate cryptocurrency time-series forecasting. First, a frequency-guided multi-scale representation module estimates dominant temporal periods from the Fourier amplitude spectrum and constructs scale-specific representations through period-dependent average pooling. This module is designed to extract multi-scale periodic information and attenuate short-term disturbances rather than to perform explicit anomaly detection. Second, the resulting representation is decomposed into trend and residual components. A lightweight linear projection is used for parsimonious trend extrapolation, while the residual component is divided into overlapping patches and processed by a Transformer encoder to model local and long-range temporal dependencies. The forecasts produced by the two branches are subsequently combined. Experiments on Bitcoin, Dogecoin, and Binance Coin data derived from the G-Research Crypto Forecasting dataset evaluate the model under multiple forecasting horizons. Comparisons with recurrent, decomposition-based, patch-based, inverted-Transformer, and multi-scale forecasting models, together with component ablations and computational-complexity analysis, are used to assess its effectiveness. The results indicate that frequency-guided multi-scale representation, decomposition, and patch tokenization provide complementary benefits for multivariate cryptocurrency forecasting. Nevertheless, the proposed frequency-guided smoothing operation does not explicitly identify statistical anomalies, and abrupt market movements may contain predictive information rather than noise.
This study proposes a Bitcoin price prediction model utilizing Long Short-Term Memory (LSTM) networks, integrating technical indicators, Reddit sentiment indicators, and on-chain data. The cryptocurrency market, particularly Bitcoin, exhibits extreme price volatility, despite its high profit potential. This volatility stems from a combination of macroeconomic factors, market participant sentiment, and fluctuations in supply and demand within the blockchain ecosystem. Existing literature typically examines only one or two types of data—whether technical, sentiment, or on-chain—without systematically verifying the complementary effects of integrating these heterogeneous data sources on predictive performance. To address this gap, this study quantitatively analyzes the contribution of each data type by constructing four experimental settings combining technical indicators, Reddit sentiment, and on-chain data based on Bitcoin price movements. Moreover, the proposed LSTM model is benchmarked against Linear Regression (LR), Random Forest (RF), and XGBoost (XGB) under identical experimental conditions. Evaluation metrics, including RMSE, MAE, and MAPE, indicate that while the LSTM model demonstrates superior predictive performance using only technical indicators, the inclusion of both Reddit sentiment and on-chain indicators results in a slight increase in error metrics. Nevertheless, this study emphasizes the potential of capturing the multifaceted characteristics of the market, which are often overlooked with single price-based indicators. It provides an empirical foundation supporting the effectiveness of heterogeneous data integration in future cryptocurrency price prediction research.
This study proposes a framework combining Variational Mode Decomposition (VMD) with a relevance-driven selection process to reduce noise and redundancy in financial time-series forecasting. The original time series is decomposed by VMD into intrinsic mode functions (IMFs), which are then evaluated using three relevance metrics: relative energy contribution, mutual information, and Spearman's rank correlation coefficient. These metrics identify the IMFs most strongly associated with future price movements. As opposed to conventional VMD-based approaches that treat all IMFs equally, the proposed relevance-driven selection process adapts IMF selection to the statistical properties of the analyzed market, thereby improving model generalization across different volatility conditions and forecasting horizons. This study makes three main contributions: (i) developing a relevance-driven IMF selection strategy to overcome limitations of traditional VMD methods, (ii) designing a hybrid framework that integrates multiscale decomposition with nonlinear information filtering, and (iii) conducting a comprehensive empirical evaluation of the proposed models. Experiments on hourly Bitcoin (BTC)/USD data from 2018 to 2025 show that the VMD-RDIC-deep learning models achieves strong forecasting performance. The results show that the proposed relevance-driven decomposition framework improves prediction accuracy and robustness compared with traditional statistical models, including Autoregressive Integrated Moving Average (ARIMA), as well as machine learning and deep learning approaches, highlighting its suitability for complex and volatile financial markets. Received: 18 January 2026 | Revised: 13 April 2026 | Accepted: 23 July 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available at https://www.kaggle.com/datasets/novandraanugrah/bitcoin-historical-datasets-2018-2024. Author Contribution Statement Maryam Maatallah: Conceptualization, Methodology, Software, Data curation, Writing – original draft, Visualization. Mourad Fariss: Software, Formal analysis, Writing – original draft. Hakima Asaidi: Validation, Investigation, Writing – review & editing. Mohamed Bellouki: Resources, Writing – review & editing, Supervision, Project administration.
A One-Byte, Options-Free Market-State Monitor: Detection-Preserving Compression of Financial Data Streams with a Class-Discriminant Token Randolph James Ferlic, M.D., and Kimberly Kate Ferlic — Fieldstone Analytics, LLC Correspondence: randolphf@fieldstoneanalyticsllc.com Preprint · Zenodo DOI: 10.5281/zenodo.22101085 · CC-BY 4.0 Abstract A previously described class-discriminant encoder reduces a window of a multivariate stream to a single 8-bit token (statistical features projected onto a shrinkage-regularized linear-discriminant and principal-component subspace, quantized to a k-means centroid, read by a lightweight head), trading thousands-fold compression for a preserved decision. We ask, under strict pre-registration, how much of that property survives on financial data — the hardest domain for naive machine learning. Across five asset classes (equities, foreign exchange, rates, cryptocurrency, commodities) and thousands of trading days of real daily price/volume data, we find a sharp, consistent boundary. As a detector, the token is near-lossless: an unsupervised codebook fit on calm windows only flags market-stress windows by nearest-centroid distance at a mean AUC of about 0.90, within 0.01–0.04 of a full 50-feature detector, at roughly 500× compression and with no labels. Pooled across a thirteen-name basket, a one-byte-per-instrument, options-free Token Market-State Index tracks the VIX volatility index (Spearman 0.62) and detects market stress that neither a full-feature detector nor VIX statistically outperforms at this sample size — a result that survives purged, embargoed walk-forward validation and block-bootstrap confidence intervals across 2010–2026, and generalizes across all five asset classes, to intraday (hourly) frequency, and to a distinct cross-asset macro risk-off state. A second, complementary detector built from the same tokens — the cross-sectional co-movement of the per-instrument token distances — flags correlation and contagion regimes, adding information beyond volatility (a +0.11 AUC point-estimate increment) and rivaling the absorption-ratio systemic-risk measure computed from the full return covariance. As a classifier or forecaster, the same token is honestly limited: it pays a real 0.07–0.16 AUC tax on supervised volatility-regime and market-state classification (only partly recovered by multi-token product quantization), it is coincident rather than leading, and it shows no directional-return skill at any horizon. The unifying regularity is that the encoder retains what a detector needs and discards what a classifier or forecaster needs — the same compression boundary observed for physiological and industrial signals, now mapped in finance. We report all negatives, including two pre-registered red flags that caught bugs in our own code before they became false results. Highlights · The single 8-bit token is a strong coincident detector of market stress: an unsupervised calm-fit codebook flags stress windows at a mean AUC of ~0.90, within 0.01–0.04 of a full 50-feature detector, at ~500× — indeed as few as two bits — compression, with no labels. · A pooled, options-free, one-byte-per-instrument Token Market-State Index tracks VIX (Spearman 0.62) and detects market stress that neither a full-feature detector nor VIX statistically outperforms at this sample size — surviving purged/embargoed walk-forward and block-bootstrap intervals across 2010–2026, and generalizing across equities, FX, rates, cryptocurrency and commodities, to intraday frequency, and to a distinct macro risk-off state. · A second detector from the same tokens — cross-sectional co-movement — flags contagion / correlation regimes, adding information beyond volatility (a +0.11 AUC point-estimate increment) and rivaling the absorption-ratio systemic-risk measure, from compressed tokens rather than the full covariance. · Honest limits, fully reported: the token pays a real 0.07–0.16 AUC classification tax, is coincident, not leading, and shows no directional-return skill at any horizon; a coincident de-risking rule reduces drawdown but is presented explicitly not as a trading strategy. Two self-caught bugs (a lookahead label and a zeroed factor) are disclosed. · No new method is claimed: the contribution is a pre-registered map of where an extreme-compression class-discriminant token is a detector and where it is not, on real public market data, with all outcomes — including those that missed their bands — reported verbatim. Statistical candor is explicit: only the market-state result carries bootstrap confidence intervals; the other comparative figures are single-split point estimates, pre-registered but multiplicity-uncorrected. What this record contains · The manuscript (PDF), with a fourteen-study summary table and eight figures. · A reproducibility archive (`PAPER_39_ZENODO_ARCHIVE.zip`): the eight pre-registration scopes with frozen outcome bands, the deterministic per-study runners (FIN-1 through FIN-14), the per-study JSON result summaries behind every figure and table value, and the figures. All data are public daily and hourly OHLCV series and the CBOE Volatility Index (VIX); no raw data is redistributed — the loaders fetch the public source at run time. Cite as R. J. Ferlic and K. K. Ferlic, "A one-byte, options-free market-state monitor: detection-preserving compression of financial data streams with a class-discriminant token," Zenodo, 2026, doi: 10.5281/zenodo.22101085. License and patent notice Released under CC-BY 4.0. Consistent with Section 2(b) of that license, no patent, patent-application, or other intellectual-property right of the authors is licensed, waived, granted, or otherwise conveyed by this publication or by any reuse of it; the methods described herein — including the single-token class-discriminant encoder, its multi-token product-quantization variant, and its unsupervised nearest-centroid-distance monitoring mode — are the subject of filed and pending U.S. patent applications, and attribution under CC-BY does not extend to those rights. Inquiries regarding licensing of the encoding methods may be directed to randolphf@fieldstoneanalyticsllc.com. Companion deposits Part of the single-token class-discriminant codebook family on Zenodo (community: spiral-domain-encoder-campaign), which includes the single-token industrial sensor substrate (doi: 10.5281/zenodo.20854722), the hardening-and-generality characterization (doi: 10.5281/zenodo.20802759), and the deterministic multi-token token-ladder (doi: 10.5281/zenodo.22003179). This deposit applies the same previously described encoding method to a new input domain — financial data streams. Keywords decision-preserving compression, class-discriminant codebook, market-stress detection, contagion and correlation regime, systemic risk, volatility regime, VIX, anomaly detection, edge computing, pre-registration, honest negatives.
Tokenized real-world assets trade continuously on public blockchains, but thevenues that price their underlyings do not. For roughly two-thirds of wall-clocktime, an on-chain protocol must value collateral against a market that is shut. This record accompanies "Coverage Inversion: Calibration-Transparent Fair-ValueOracles for Closed-Market Hours". The paper inverts the conventionalpoint-plus-confidence oracle interface: the target coverage level tau becomes apublished product input, and every served price band carries a calibrationreceipt that a third party can reconstruct from public data. CONTENTS The paper (67 pages) and the LaTeX source arXiv compiles. A reference implementation in three languages — the Python serving path, its Rust port (pinned to the Python by a 329-case parity harness), and the Anchor programs for the on-chain publish path. The calibration artefacts: the 20 deployment scalars that define the served bands, including the SHA-256-stamped frozen artefact used for out-of-sample validation. The public band archive: an append-only record of bands actually served, with Saturday width commitments published before Monday's open, so the claims can be audited after the fact rather than taken on trust. EVIDENCE Two closed-market panels over the same ten US-listed tickers, 2014-2026: 5,996 weekend windows (Friday close to Monday open) and 22,624 overnight windows (close to next open). The headline weekend result is held out by leave-one-symbol-out cross-validation at tau = 0.95: realised coverage 0.9497 +/- 0.0128, every fold passing Kupiec. On a 40-cell symbol-by-tau grid the architecture passes 40 of 40 Kupiec tests, against 31 of 40 for the strongest practitioner baseline (GARCH-t). The same architecture, with only its gap selector changed, carries from weekends to overnight gaps — calibration transparency is a property of closed market hours generally, not of weekends specifically. Because an earnings release has a publicly known date and session, the band widens deterministically ahead of it. No incumbent oracle exposes calendar-conditioned coverage. LICENSING This record is CC BY 4.0. The source code in the reference-implementation archive is Apache-2.0 and ships its own LICENSE file, which governs that archive.
The rapid expansion of the digital currency market and the growing role of stablecoins as potential intermediaries have brought its interconnectedness with traditional financial markets to the forefront of global financial research. Using daily data from 4 January 2021 to 30 September 2025, this study constructs a variable system with the price indices of USDT and USDC as core digital currency proxies, alongside traditional financial asset indices for stocks, bonds, and gold derived via the entropy weight method. We employ a comprehensive set of econometric techniques, including static correlation analysis, vector autoregression (VAR), impulse response functions, and extreme-event shock tests, to systematically investigate the interdependence structure, risk spillover dynamics, time-varying co-movements, and structural changes between the two markets during extreme risk episodes. The findings reveal an overall weak and asymmetric bidirectional spillover relationship between the cryptocurrency and traditional financial markets. Volatility in the digital currency market is found to be largely endogenous, with a limited capacity to transmit shocks externally. Conversely, traditional financial markets—particularly the equity market—exert a more pronounced influence on the digital currency market. Critically, under the impact of extreme risk events, the cross-market linkages exhibit structural breaks; the direction and intensity of correlation can strengthen significantly or even reverse, demonstrating a clear state-dependency. This research provides empirical evidence for understanding the functional role of digital assets within the macro-financial system, their risk transmission pathways, and their implications for systemic financial stability. The findings offer valuable theoretical and practical insights for financial regulators in designing robust cross-market risk prevention frameworks and for investors seeking to optimize asset allocation strategies.
Dr. Mohamed Nachat, Hassan Oukhouya, Saïd El Melhaoui, Moustapha Faizi · 7 authors
Time series forecasting plays a central role in finance, energy, and public health. Classical statistical, machine learning, deep learning, and generative approaches have all been applied to forecasting tasks in these fields, but comparisons between them are usually confined to a single domain or to models from the same family, and few studies report both univariate and multivariate results under the same conditions. This paper presents a controlled cross-domain comparison of four representative paradigms: classical statistics (Seasonal Autoregressive Integrated Moving Average with Exogenous variables, SARIMAX), gradient boosting machine learning (Light Gradient Boosting Machine, LightGBM), recurrent deep learning (Recurrent Neural Network, RNN), and generative-adversarial deep learning (Conditional Generative Adversarial Network, CGAN). Each model is evaluated on three monthly datasets with contrasting characteristics: Bitcoin prices (175 observations, high volatility), U.S. energy consumption (612 observations, strong seasonality), and U.S. cardiovascular mortality (300 observations, gradual trend with pandemic shock). Both univariate and multivariate variants are tested under the same preprocessing and one-step-ahead evaluation protocols, using eight performance metrics. The CGAN reaches the lowest MAPE on energy consumption (2.88%). On Bitcoin, the multivariate LightGBM lowers the MAPE from 28.26 to 19.25%, while on cardiovascular mortality the RNN reaches 3.34% MAPE. No paradigm performs best in every domain, and the gain from exogenous variables depends on both the paradigm and the domain.
Background. Technical analysis remains among the most accessible forms of market decision support, but its incremental predictive value depends on design choices, and repeated model selection can create spurious backtest performance. Objective. This study evaluates whether widely used technical indicators, reconstructed from public formulas, can predict the next daily price direction of Bitcoin, the S&P 500, and gold. Methods. Daily data from 1 January 2015 to 14 July 2026 are examined using a chronological walk-forward design. Every signal observed at the close of day t is matched only with the sign of the close-to-close return from t to t+1. The main out-of-sample period begins in 2019, with 2019–2022 used for model selection and 2023–2026 reserved for confirmation. Performance is measured primarily by balanced accuracy and supplemented by accuracy, directional recall, stationary-bootstrap confidence intervals, and after-cost trading outcomes. Results. The best single indicator, Ichimoku 9/26/52, produced a macro balanced accuracy of 50.9%, while the best predetermined combination, Volume Confirmed, reached 50.8%. A new ridge-logistic hybrid indicator, SPAH-1, achieved 51.9% in validation and 51.3% in confirmation. Its confirmation balanced accuracy was 49.8% for Bitcoin, 49.2% for the S&P 500 proxy, and 55.0% for gold; only gold's bootstrap interval excluded 50%, but its predictions were strongly biased toward the upward class. After trading costs, Volume Confirmed underperformed buy-and-hold for all three assets. Additional selective experiments did not support an 80% daily prediction target. Conclusion. Technical indicators may assist regime description and decision confirmation, but they do not provide a robust universal next-day forecasting edge.
Harini D., S. Aruna, Santhanalakshmi V., D. P. Sivasakti Balan · 5 authors
Cryptocurrencies have emerged as a significant component of the global financial systems, attracting considerable attentions from investors, researchers and policymakers. Simultaneously, Indian sectoral stock market plays a crucial role in the country’s economic development and investment landscape. Due to their high volatility, crypto currency markets may influence traditional financial markets and investment decisions .This study investigates the impact of crypto currency market movements from on selected Indian sectoral stock markets, including the information Technology, Banking, Pharmaceutical sectors. Historical data from major cryptocurrencies and sectoral stock indices like Bitcoin, Ethereum, Nifty IT, Nifty Bank, and Nifty Pharma indices which was collected and analyzed using explainable regression techniques. Based on the given datasets, the findings provide valuable insights into the relationship between cryptocurrency markets and selected Indian sectoral stock market indices, particularly in understanding how crypto currency market fluctuations during major global events, such as pandemics, geopolitical conflicts, and economic uncertainty, may influence investor behavior and sectoral performance across the Indian economy.
The paper presents a new hybrid approach (LSTM-RF-SLSQP) that acts as an advanced decision-support tool for institutional crypto portfolio management. The methodology includes the use of LSTM neural networks to detect non-linear temporal patterns and RF algorithms to detect structural market noise. The ability to predict future prices based on LSTM-RF model has been extensively verified out-of-sample using Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), and shown to have smaller prediction error than standalone algorithms on very volatile assets. Then, based on robust return and risk forecasts, SLSQP optimization algorithm allocates asset weights aiming at maximizing Sharpe ratio under specific institutional constraints, applying buy-and-hold approach with quarterly rebalance. The empirical study is performed for a portfolio of ten major cryptocurrencies (BTC, ETH, SOL, BNB, XRP, ADA, DOGE, TRX, LINK, DOT) using data provided by KuCoin exchange from January 2024 to January 2026. The numerical experiments demonstrated the significant superiority of the suggested framework compared to all benchmarks. Namely, the LSTM-RF-SLSQP approach provides the impressive annual return of 107.40%, Sharpe ratio of 2.04, with maximum drawdown equal to -3.80%.
A fixed-odds contract on short-horizon price direction has a positive expected value only when its win probability exceeds the break-even rate implied by the payout ratio. A deployable predictor must also produce signals at a sufficiently stable rate. We formulate this setting as selective prediction with a coverage constraint and combine a five-seed gradient-boosting ensemble over a 90-dimensional causal feature panel with daily adaptive quantile thresholds, each estimated from the preceding 14 to 28 days of model scores, with parameters selected on training data alone. Configurations are frozen after three chronological pseudo-out-of-sample folds and evaluated on a held-out period from 1 January to 10 June 2026, and the whole procedure is then repeated on a quarterly re-freezing cadence over seven successive windows. Across BTC and ETH at 5- and 10-min horizons, with a payout of 0.8 and a 55.56% break-even rate, the models execute 10.4 to 11.0 trades per day, and all four selective win rates exceed break-even. Under a dependence-aware block bootstrap, three of four remain significant, and within a 32-test confirmatory family, two survive Holm–Bonferroni correction. Coverage stays inside the operational band in 26 of 28 re-frozen windows. Compared under one execution protocol, a fixed calibration slice drifts out of band while a trailing window does not, and adaptive conformal inference (ACI) matches the proposed rule on coverage when its step size is tuned but not otherwise, whereas an outcome-driven conformal controller reduces coverage by more than an order of magnitude. The expected value is insensitive to exchange fees, which consume under 5% of the measured edge, and sensitive to the payout term. Under matched feature sets, training pools, and coverage, most of the apparent cross-asset difference does not persist. This paper presents a proof of concept for the framework rather than making any claim about cryptocurrency predictability.
Although stablecoins occupy a segment of digital-asset markets in which price stability is central by design, their temporary departures from reference values may reveal important information about latent risk and market stress. In this paper, we examine whether bubble and crash signals extracted from traditional cryptocurrencies and stablecoins improve volatility, Value-at-Risk, and Expected Shortfall forecasting and, in connection with these forecasting gains, contribute to the assessment of cross-asset contagions. The analysis applies the Bubble Crash–GARCH models, in which extreme price phases are identified through the Phillips, Shi, and Yu real-time monitoring procedure and incorporated into the conditional mean of returns through event-based dummy variables. For stablecoins, extreme episodes are not inferred from price dynamics in isolation but from deviations between the observed price and the asset-specific reference value. The empirical investigation focuses on Bitcoin, Ethereum, Tether’s USD-pegged (USDT), and Tether Gold and evaluates asset-specific bubble–crash effects and bidirectional contagion channels between traditional cryptocurrencies and stablecoins, using Bitcoin and Tether as the leading representatives of the two market segments. The findings indicate that accounting for bubble and crash episodes leads to more accurate volatility forecasts than standard GARCH benchmarks. For Value-at-Risk and Expected Shortfall, the bubble–crash specifications can improve tail risk forecasting at several tail probability levels through more accurate coverage, lower quantile loss, and stronger ESR backtesting performance. The results also reveal different degrees of price exuberance across the two asset categories: while extreme price dynamics are more evident among traditional cryptocurrencies, deviations from fundamentals are rare for stablecoins. Among stablecoins, USDT exhibits limited but detectable exuberance, whereas Tether Gold does not display extreme price episodes. However, when such deviations occur, as in the case of USDT, they generate significant contagion effects on major cryptocurrencies. Notably, extreme episodes originating in USDT have a stronger impacts on Bitcoin and Ethereum than the reverse spillovers from traditional cryptocurrencies to USDT. Overall, the evidence suggests that stablecoins are not merely passive instruments within the digital-asset ecosystem. Even temporary deviations from their reference values contain valuable information for risk forecasting and contagion monitoring.
Large language models (LLMs) have shown strong potential for financial analysis and trading, but direct trading remains challenging because the predictive capabilities required can vary across assets, decision fields, and market conditions. Existing LLM-based trading systems either coordinate human-defined external experts or adopt conventional internal Mixture-of-Experts (MoE) routers that do not directly evaluate how individual experts contribute to trading decisions. Moreover, these routers receive no direct signal indicating when an inactive expert has become more suitable as market conditions change. We find that native router scores poorly reflect how much individual experts improve trading decisions, frequently leaving better alternatives unselected. We further reveal that token-specific expert usefulness exhibits a compact low-dimensional structure. Based on these findings, we propose TradingMoE, a trading-oriented sparse MoE that augments a frozen dense LLM with lightweight residual experts. We introduce a Query-Key router that represents the expertise required by each token under the current market context as a low-dimensional query and matches it with learnable expert keys. We further propose a sparse expert selection update mechanism that samples a few inactive experts during training and estimates whether they should replace the weakest expert in the current Top-k route. This mechanism enables the router to update expert selection as market conditions change while preserving sparse computation. Experiments against 22 baselines on stock and cryptocurrency markets show that TradingMoE improves cumulative return over the best-performing baselines by 30.89% and 30.7%, respectively. Rolling paper-trading experiments further demonstrate that its advantage persists under forward-only deployment.
The programmable economy lacks a universal computational layer capable of interpreting, translating, verifying, and simulating the mathematical and cryptographic operations that underpin digital assets. Existing tools are fragmented: wallet software provides only rudimentary transaction signing, portfolio trackers offer aggregated views without evidence, and specialized calculators address isolated problems. No general-purpose, cryptographically verifiable, language-native computational environment exists for digital value. KHOTOR is designed to fill this gap. It is a universal, deterministic runtime that interprets the anti-entropic linguistic protocol Kryptophon, transforms plain-language queries into executable computational expressions, and performs multi-domain financial mathematics across asset conversion, transaction analysis, decentralized finance, tokenomics simulation, cryptographic proof generation, and risk assessment. Every output carries an epistemic classification — verified, observed, inferred, simulated, or uncertain — and can be exported as a Gamma-Proof: a cryptographically signed, independently verifiable artifact. This paper presents the complete KHOTOR architecture: a ten-layer computational engine, a formal abstract machine for Kryptophon evaluation, a tiered adoption model that makes the programmable economy accessible to non-technical users while creating a new domain of expertise for professionals, and a product family spanning a public cloud API, a web platform, a handheld consumer device, and integration with dedicated hardware instruments. All components are designed around a single governing principle: every calculation shows its work, every output carries a truth label, and no inference is ever presented as fact.
Cryptocurrency time-series forecasting is a challenging task because market data usually exhibit high noise, strong volatility, non-stationarity, nonlinear dynamics, and long-range dependencies. In addition, multivariate market indicators often contain redundant or weakly informative variables, which may reduce forecasting accuracy and model interpretability. To address these issues, this study proposes BSFinformer, a Boruta-SHAP enhanced Finformer framework for multivariate cryptocurrency time-series forecasting. The proposed framework first applies a leakage-aware Boruta-SHAP feature selection strategy to identify informative market variables and remove redundant features. To avoid temporal information leakage, feature selection is performed only on the training set, and the selected feature subset is then applied unchanged to the validation and test sets. The selected features are subsequently fed into an improved Finformer model that integrates temporal embedding, sequence decomposition, and sparse self-attention to capture local fluctuations, trend evolution, and long-range temporal dependencies. Experiments are conducted on three cryptocurrency assets, namely Bitcoin, Dogecoin, and Binance Coin, using chronological train–validation–test splits. The proposed model is compared with classical forecasting models and recent long-sequence forecasting baselines, including LSTM, Transformer, Informer, Autoformer, DLinear, PatchTST, TimesNet, and iTransformer. Experimental results show that BSFinformer achieves competitive forecasting performance in terms of MSE and MAE. Ablation experiments further demonstrate the contributions of Boruta-SHAP feature selection, temporal embedding, sequence decomposition, and sparse self-attention. These results indicate that feature-selected temporal modeling can improve forecasting accuracy and interpretability for multivariate cryptocurrency market data.
The fluctuation characteristics of financial time series have always been one of the research hotspots in the academic community. Generally speaking, financial return series have the characteristics of volatility clustering, fat tails, conditional heteroskedasticity, asymmetric shocks, etc. The above phenomena can be explained from the perspective of dynamic conditional variance by GARCH models and their extensions. This paper first introduces the basic ideas of ARCH and GARCH models, with a focus on the issue of volatility clustering of financial returns. Then, it reviews the relevant research from three aspects: model evolution, application scenarios, and practical value. It also analyzes the role of GARCH-type models in capturing volatility persistence, asymmetric impact, and risk transmission through applications in cryptocurrencies, energy assets, and high-frequency financial data. The study shows that GARCH-type models capture the volatility clustering feature of financial returns well, but there is still room to improve the modeling of extreme risk, the handling of high-dimensional assets, and model interpretability.
Volatility control converts risk estimates into portfolio exposure, yet existing approaches often rely on a fixed volatility estimator or a pre-defined control rule that may not adapt to changing market conditions. We propose VolRouter, a modular framework that formulates volatility control as state-conditioned routing over estimator-controller pairs. VolRouter first summarizes market conditions into a control-relevant state profile and then performs routing through three stages: state inference, switch review, and pair selection. The Router can be implemented using rule-based, learnable, or LLM-based decision modules, while portfolio actions remain generated by predefined control policies. We evaluate VolRouter across S&P 500, Multi-Asset, Bitcoin, and USDT volatility-control settings. VolRouter achieves the highest Sharpe ratio in three of four settings. On S&P 500, it improves Sharpe from 0.952 for RV + Naive Scaling to 1.222 while reducing maximum drawdown from 15.10% to 12.58% and daily CVaR from 1.76% to 1.32%. On Multi-Asset, it improves Sharpe from 1.498 to 1.540 and reduces CVaR from 1.56% to 1.18%. Bitcoin shows similar improvements in risk-adjusted performance, while USDT provides a boundary case where simpler state-aware selectors remain competitive. Ablation and sensitivity analyses show that the improvement comes from relative policy evaluation and selective persistent switching rather than simply expanding the policy library. These results suggest that volatility control can be viewed as a policy-selection problem when risk management requirements vary across market states.
Velian Prapatoni, Rizky Parlika, Firza Prima Aditiawan
Cryptocurrency markets are characterized by high volatility, rapid price fluctuations, and substantial uncertainty, creating challenges for investment risk interpretation. This study develops a descriptive risk-interpretation framework, rather than a price-prediction or decision-optimization model, by integrating multi-scale volatility analysis with technical indicators. A quantitative descriptive design was applied to approximately one year of historical hourly price data for Bitcoin and Ethereum, covering open, high, low, close, volume, and percentage change attributes. The data were chronologically sorted, numerically cleaned and normalized, transformed into log returns, and analyzed through rolling standard deviation. Volatility was estimated across three explicitly defined horizons: short-term 7-period, medium-term 30-period, and long-term 90-period rolling windows. Moving Average (MA), Relative Strength Index (RSI), and Average True Range (ATR) were then incorporated to contextualize trend direction, momentum, and fluctuation intensity. The results show that volatility is strongly horizon-dependent: short-term movements responded more sharply to market shocks, whereas longer horizons produced smoother risk patterns. Across the analyzed Bitcoin and Ethereum hourly series, the reported 42.3% short-term and 21.7% medium-term increases were calculated as relative differences against long-term rolling volatility during identified high-uncertainty intervals, not as predictive accuracy measures. These findings indicate that combining rolling volatility with MA, RSI, and ATR can improve the transparency of descriptive cryptocurrency risk assessment. The framework may support preliminary interpretation for novice or risk-averse investors, although it does not empirically test investor comprehension or subsequent decision quality.