High-frequency crypto forecasting requires systems that are accurate, explainable, and designed for human decision-making. Bitcoin presents a unique challenge for Human-Centred AI (HCAI) due to its volatility and sensitivity to heterogeneous technical, fundamental, and sentiment signals. This paper presents an explainable multimodal framework for Bitcoin forecasting at 15-minute resolution. We align five modalities—market data, on-chain metrics, the Fear & Greed Index (FGI), news, and Reddit—onto a unified, leakage-safe 15-minute grid. We evaluate tree-based, sequential, and Multimodal Fusion Block (MFB) models for next-interval log-return prediction using chronological splits. Results show that while short-horizon prediction remains challenging, multimodal features consistently improve over structured baselines, particularly during event-driven periods. To ensure transparency, the framework integrates a dual-layer explanation system: SHapley Additive exPlanations (SHAP) attributions combined with large language model (LLM) narratives, ensuring outputs are both technically faithful and human-accessible. This work unlocks the “black box” of complex predictive architectures, transforming opaque multimodal signals into transparent, actionable decision support for high-frequency trading.
Abstract Automatic trading systems cope with the needs of put out emotional biases from the trading operation of public assets. These systems place orders based on a price model that forecasts the future price of an asset. Those systems, developed by edge funds and institutional investors, are not available to the public, and extensive research in this field is worth the effort. In this research, we developed a short-term price model based on a neural network and used it to forecast the near-future price direction. More in depth, we introduced the feature extraction process and parametric labeling strategy to build an ML ready dataset that includes more than 400 cryptocurrencies. The model is then validated by building a trading strategy on the two most capitalized cryptos at the time of writing: Bitcoin and Ethereum. The validation uses a trading simulation that spans six years of historical data for Bitcoin and Ethereum, including both retrospective (backtest) and prospective (forward test) evaluations. The results demonstrate that the neural network-based model exhibits a very good generalization to patterns found in historical data, enabling predictions in future data within the trading simulation. In addition, a comprehensive analysis of the importance of features was conducted to enhance the interpretability and performance of the model. Finally, we test our model in a simulated trading session; it shows that, with a simple buy-only strategy plus a stop loss, the trading system limits the draw dawn during bear markets.
This study examines the dynamic interconnections and portfolio implications of clean energy ETFs, artificial intelligence (AI) indices, crude oil, and Bitcoin within sustainable and technology-driven financial markets. Using a Time-Varying Parameter Vector Autoregression (TVP-VAR) framework and daily data from January 2019 to December 2024, we analyze time-varying spillovers and construct optimal portfolios based on dynamic connectedness measures. The results show that clean energy and AI-related assets display relatively stable portfolio weights, whereas Bitcoin exhibits highly volatile and generally limited allocations, particularly under risk-averse strategies. Conventional approaches such as the Minimum Variance and Risk Parity portfolios tend to favor traditional assets, while the Maximum Connectedness Portfolio enhances diversification by allocating more weight to weakly connected assets, including Bitcoin and green ETFs. The findings offer practical insights for resilience-oriented and innovation-driven portfolio construction.
Decentralized Finance (DeFi) has become a major component of digital asset markets, yet accurately valuing protocol performance remains difficult due to high volatility, nonlinear pricing dynamics, and persistent disclosure gaps that amplify valuation risk. This study develops an Optuna-tuned Super Learner stacked ensemble to improve risk-aware DeFi valuation, combining Extremely Randomized Trees (ETs), Support Vector Regression (SVR), and Categorical Boosting (CAT) as heterogeneous base learners, with a K-Nearest Neighbors (KNNs) meta-learner integrating their forecasts. Using an expanding-window panel time-series cross-validation design, the framework achieves significantly higher predictive accuracy than individual models, benchmark ensembles, and econometric baselines, obtaining RMSE = 0.085, MAE = 0.065, and R2 = 0.97—representing a 25–36% reduction in valuation error. Wilcoxon tests confirm that these gains are statistically significant (p < 0.01). SHAP-based interpretability analysis identifies Gross Merchandise Volume (GMV) as the primary valuation determinant, followed by Total Value Locked (TVL) and key protocol design features such as Decentralized Exchange (DEX) classification, while revenue variables and inflation contribute secondary effects. The findings demonstrate how explainable ensemble learning can strengthen valuation accuracy, reduce information-driven uncertainty, and support risk-informed decision-making for investors, analysts, developers, and policymakers operating within rapidly evolving blockchain-based digital asset environments.
This study reviews the advancements in AI-driven methods for predicting stock prices, tracing their evolution from traditional approaches to modern finance. The role of AI in the market extends beyond predictive systems to encompass the intersection of financial markets with emerging technologies, such as blockchain, and the potential influence of quantum computing on economic modeling. A decentralized finance system examines the application of Reinforcement Learning in financial market prediction, highlighting its potential for continuous learning from dynamic market conditions. The study discusses the development of hybrid prediction models, stock market machine learning systems, and AI-driven investment portfolio management. The potential of quantum computing enhances portfolio analysis, fraud detection, optimization, and asset valuation for complex market predictions, as well as the impact of blockchain technologies on transparency, security, and efficiency. Machine learning techniques can significantly automate data collection and purification. Financial decision-making and the application of time-series analysis techniques can be readily learned through deep reinforcement learning for stock price prediction. Deep Neural Networks and Strategic Asset Allocation can be managed by evaluating performance and portfolio using real-time market insights from AI models. Although there are numerous ethical, sentimental, regulatory, and data quality issues in market prediction, the future job market is heavily dependent on these criteria, particularly through effective risk management and fraud detection.
In this work, we utilize the blockchain transactions and financial instruments to pre-dict the Bitcoin price using machine learning. We use three models: Light Gradient Boosting Machine (LightGBM), Decision Tree Regressor and Random Forest Regressor applied on a feature set which includes lagged close prices, 14-day Simple Moving Av-erage (SMA), Relative Strength Index (RSI) and daily confirmed Bitcoin transactions. The data is temporally aligned and pre-processed to maintain temporal coherence, as well as for conversational fluency. Through the results assessment by means of RMSE MAE, MAPE and R², we can found that Random Forest model has results closer to best performance with values of: 264.81 (RMSE); 175.41(MAE); for MAPE is 0.27% and; R² equals to 0.9958. Our findings also lend strong support for the effectiveness of simul-taneously considering not only blockchain-specific market variables but also tradi-tional financial predictors towards improved model performance and generalization. Our findings underscore the importance of raw blockchain transaction data for pre-dicting cryptocurrency prices, and present a new tool for data-based decision making in decentralized finance.
Kassem Danach, Hassan Rkein, Ahmad Farroukh, Ziad E. L. Balaa · 5 authors
The static and hard-coded logic of smart contracts in Decentralized Finance (DeFi) platforms significantly limits their adaptability in dynamic and volatile market environments. To address this challenge, we propose a novel hyper-heuristic driven framework that enables real-time rule optimization within smart contracts, thereby enhancing responsiveness, gas efficiency, and operational robustness. The framework features a two-layer architecture: a reinforcement learning-based high-level controller selects appropriate low-level rule heuristics from a domain-specific library based on evolving transaction contexts and on-chain data. Implemented and evaluated on Uniswap v2 and Aave v3 protocols, the system dynamically optimizes parameters such as slippage tolerance, gas usage thresholds, and loan-to-value ratios. Experimental results on real-world datasets show significant performance improvements, including a 45.6% increase in transaction success rate, 28.3% reduction in average gas consumption, and 38.4% drop in liquidation events under market stress scenarios. This research demonstrates the feasibility and advantages of embedding intelligent, adaptive decision-making mechanisms within DeFi smart contracts, opening new pathways toward autonomous, resilient, and regulation-aligned blockchain systems.
Blockchain ecosystems, particularly decentralized finance (DeFi) platforms, have become high-value targets for adversaries exploiting vulnerabilities in smart contracts. Traditional honeypots designed to attract such attackers are often static and easily detectable once adversaries recognize repeating patterns. In this paper, we propose Adaptive AI-Powered Smart Contract Honeypots (AASCH), a novel deception-based security mechanism that dynamically evolves its behavior using reinforcement learning techniques. The honeypot continuously adapts to attacker strategies by modifying contract responses, transaction logic, and resource interactions in real time, thereby creating an unpredictable deception environment. We implement AASCH on the Ethereum test network and simulate various exploit scenarios, including reentrancy attacks, flash-loan exploits, and oracle manipulations. Experimental results demonstrate that AASCH significantly improves attacker capture rates (up to 37% higher than static honeypots) while reducing evasion attempts and false positives. Our findings indicate that adaptive deception is a promising defense strategy for strengthening blockchain ecosystems against evolving cyber threats.
For many decades, time-series forecasting has been applied to different problems by scientists and industries. Many models have been introduced for the purpose of forecasting. These advancements have significantly improved the accuracy and reliability of predictions, especially in complex scenarios where traditional methods struggled. As data availability continues to expand, the integration of machine learning techniques is likely to further enhance forecasting capabilities across various fields. Today, hybrid techniques are gaining popularity, as they combine the advantages of different approaches to deliver improved predictive performance and more advanced visualization analytics for decision support. These hybrid approaches can provide better prediction, and at the same time, they can develop a more sophisticated set of visualization analytics for decision support. Recently, the integration of cross-entropy, fuzzy logic, and attention mechanisms in hybrid forecasting models has enhanced their ability to capture complex and uncertain patterns in financial and energy markets. In this study, we propose a hybrid ANN–LSTM deep learning model optimized with cross-entropy, fuzzy logic, and an attention mechanism to enhance the forecasting of financial and energy time series, specifically Ethereum and natural gas prices. Our models combine the feature extraction strength of ANN with the temporal learning of LSTM, while cross-entropy improves convergence, fuzzy logic handles uncertainty, and attention refines feature weighting. Since inaccurate forecasts can lead to greater estimation uncertainty and increased financial and operational risk, improving predictive reliability is essential for effective risk mitigation. These techniques prove effective not only in improving estimation accuracy but also in minimizing financial risks and supporting more informed investment decisions.
While deep learning models have demonstrated superior performance in cryptocurrency forecasting, their deployment is often hindered by a lack of interpretability and trustworthiness. To bridge this gap, this paper proposes the Cryptocurrency Counterfactual Explanation (CryptoForecastCF) model. Recognizing the inherent volatility and complex non-linear dynamics of cryptocurrency markets, we argue that understanding the sensitivity of model outputs to slight variations in historical conditions is fundamental to robust risk management. CryptoForecastCF employs a gradient-based optimization strategy to generate meaningful counterfactual explanations. Specifically, it identifies minimal modifications, defined as the optimal perturbations to historical market features such as price constrained by ℓ1 or ℓ2 norms, that are sufficient to steer the model's future predictions into user-specified target intervals. This approach not only elucidates the key driving factors and decision boundaries of opaque models but also equips traders and risk managers with actionable insights, enabling them to identify the specific market shifts required to navigate high-stakes scenarios and mitigate unfavorable predictive outcomes.
Cryptocurrency price forecasting has gained increasing attention due to the market’s high volatility and structural complexity. While many recent studies have explored deep learning architectures, including attention- and transformer-based models, existing research still faces notable limitations: (i) inconsistent feature engineering choices, (ii) limited examination of hybrid machine-learning models, and (iii) a lack of transparent trading evaluation using realistic backtesting assumptions. To address these gaps, this study develops a hybrid forecasting and trading framework based on Support Vector Regression (SVR) combined with a set of rule-based technical strategies. Using four major cryptocurrencies – BTC, ETH, XRP, and LTC – from 2018 to 2020, the proposed framework integrates thirteen technical indicators with a sliding-window scheme and compares SVR against Random Forest (RF) and Long Short-Term Memory (LSTM) benchmarks. Empirical results show that SVR offers a competitive balance between predictive accuracy and computational efficiency, particularly in moderate-volatility regimes. The strategy backtesting further demonstrates that SVR-driven signals can outperform traditional technical rules under certain market conditions, although limitations remain for highly volatile assets such as Bitcoin. The study contributes to the literature by clarifying feature-design choices, evaluating SVR within a multi-asset setting, and providing reproducible code and datasets through an open-access repository.
Prediction markets have grown from niche research focused platforms into a multi-billion-dollar industry, yet the academic literature on the properties of these new platforms has not caught up. Prediction markets on the touch probability of traded assets are a new form of contract, in which traders bet on whether an asset will reach a specified price level within a given window. They represent a structurally distinct and largely unexamined category. Because these contracts reference a continuously traded underlying with a deep options market, the probability they quote may already be embedded in existing market data. This thesis asks if Polymarket Bitcoin price-target contracts contribute to price discovery, or whether they merely repackage the touch probability already implied by the Bitcoin spot and options markets. The analysis uses 258 monthly barrier contracts traded between October 2024 and March 2026 and proceeds in three stages: a Mincer-Zarnowitz calibration regression, per-contract Granger causality tests at the hourly frequency, and a cross-sectional comparison of Polymarket prices against a closed-form one-touch barrier probability evaluated under both implied and realised volatility. The calibration test cannot reject Polymarket prices are unbiased probability forecasts of their resolution outcomes. The Granger causality tests reveal that Polymarket no detectable predictive content for next-hour Bitcoin returns. A closed-form barrier probability computed from the spot price, strike, time to maturity, and a volatility estimate explains over 95% of the cross-sectional variation in Polymarket prices under both implied and realised volatility, though the joint null of exact equality is rejected. The price tracks the realised-volatility benchmark and sits below the implied-volatility one, so the level departure is consistent the variance risk premium embedded in options rather than independent information about Bitcoin. The three results replicate out of sample on 252 Ethereum contracts. Taken together, the findings suggest that financial-asset prediction markets contribute little information beyond what is already priced in the underlying spot and options markets.
This study investigates the existence of long-run relationships between cryptocurrency prices (Bitcoin, Ethereum) and macro-economic and macro-financial variables, addressing a gap in prior research primarily focused on short-run responses. Using time series data on these variables over the period January 2022 to December 2024, statistical co-movement tests are applied to find significant relationships. Tests conducted with monthly data reveal significant co-movement in Bitcoin and Ethereum prices and the Consumer Sentiment Index, global gold reserves (measured in ounces), the MSCI World Index, and the Producer Price Index. Additionally, weekly tests reveal co-movement between Ethereum and gold prices in calendar year 2023. These findings provide empirical evidence that Bitcoin and Ethereum increasingly reflect certain macro-financial and macro-economic conditions rather than trading independently of traditional economic forces, while evidence supporting a stable “digital gold” role remains sparse and episodic.
This paper examines the return connectedness between Bitcoin and stock indices of economies with high levels of cryptocurrency adoption. Such economies are predominantly emerging markets characterized by elevated inflation, poor institutional quality, and macroeconomic and political instability, creating conditions under which investors may reallocate from traditional assets to Bitcoin during episodes of increased uncertainty. To assess this linkage, we employ a TVP-VAR framework with frequency-domain decomposition. Our results indicate only modest return connectedness under normal market conditions, which intensifies during periods of market turmoil. This observed pattern, along with low correlation and the identification of Bitcoin as a net return receiver, led to testing the portfolio diversification potential of Bitcoin. The evidence indicates that Bitcoin contributes to both risk mitigation and return enhancement at low hedging costs. The effect is more pronounced for emerging-market portfolios than for developed markets.
En menos de un año, Bolivia ha pasado de prohibir el uso de criptomonedas a incorporar su uso de manera frecuente. Esta investigación examina la dinámica de los retornos y el riesgo asociado al Bitcoin, la criptomoneda de mayor valor en el ecosistema cripto, mediante modelos diseñados para activos de alta volatilidad. El análisis se basa en una serie temporal de datos diarios recopilados durante doce años, con énfasis en la medición de retornos negativos. Los resultados muestran que la media de los retornos es positiva y estadísticamente significativa, aunque su capacidad explicativa sobre la variabilidad total es limitada, lo cual es consistente con el comportamiento típico de series financieras de alta frecuencia. En cuanto a la volatilidad, se confirma la presencia de heterocedasticidad condicional, con efectos ARCH y GARCH altamente significativos. La persistencia de la volatilidad, evidenciada por un coeficiente GARCH cercano a uno, indica que los episodios de alta o baja volatilidad tienden a mantenerse en el tiempo. Estos hallazgos destacan la relevancia de modelar adecuadamente la varianza condicional en el análisis de activos financieros como el Bitcoin. Adicionalmente, se identificó la necesidad de ajustar la escala de los datos, recomendándose una rescalación previa para mejorar la precisión en futuras estimaciones.
This work deals with the possibility of using recurrent neural networks of the LSTM and GRU type for predicting daily logarithmic returns of selected cryptocurrencies (BTC, ETH, LTC, BNB). Based on daily OHLCV data from the period 2017–2024, three sets of input variables are constructed: a basic set (transformed price and volume variables and logarithmic returns), a set of technical indicators, and a set of technical indicators together with macroeconomic data. The LSTM and GRU prediction models are calibrated for different memory lengths (10, 20, 50 days), numbers of neurons, and all three sets, and their performance is evaluated using RMSE, MAE, and directional accuracy. The results show that daily returns of cryptocurrencies are difficult to predict from the point of view of a one-time prediction: random walk remains a very strong benchmark in all cases, and the best neural networks only come close to it. Subsequently, simple Long/Short strategies are constructed based on the predictions and compared with the passive Buy&hold strategy. For all four cryptocurrencies, configurations are found whose strategies achieve higher annualized returns and Sharpe ratios than Buy&hold in the test period, especially for more volatile altcoins. However, this outperformance is conditioned by ex post selection of the “best” models and neglect of transaction costs, and therefore it must be interpreted with caution as an illustration of the potential and limits of deep neural networks in short-term prediction of cryptocurrencies.
The rapid expansion of decentralized finance (DeFi) has generated rich, transparent on-chain data that remains largely underutilized in high-frequency trading models. Most existing studies rely primarily on centralized exchange (CEX) price feeds, which often suffer from low signal-to-noise ratios (Lim et al., 2021; Lee et al., 2025). This study develops a multi-modal forecasting framework for the WETH/USDC 0.05% fee tier pool on Uniswap v3. We integrate Binance CEX market microstructure data with granular onchain DEX metrics, including swap imbalance, on-chain volume, liquidity depth, tick velocity, and pool liquidity utilization. An XGBoost classifier was trained on synchronized 15-minute bars, with realistic cost-aware backtesting incorporating pool fees and slippage. The model achieved a directional accuracy of 63.26% on out-of-sample data. Feature importance analysis revealed that on-chain variables, particularly volume_usd, liquidity_usd, imbalance, and tick_velocity, ranked among the top predictors. In cost-aware backtesting, the strategy outperformed Buy & Hold by 2.49 percentage points, although absolute returns remained modestly negative due to transaction costs. This research demonstrates the incremental predictive value of integrating Uniswap v3 on-chain DEX metrics with CEX data for high-frequency forecasting. While transaction costs remain a significant challenge, the findings highlight the potential of multi-modal approaches in DeFi markets and provide a foundation for future work using more advanced architectures such as the Temporal Fusion Transformer..
This thesis investigates whether the integration of digital assets, specifically Bitcoin and <br/>Ethereum, into a traditional multi-asset institutional portfolio can produce superior out-of<br/>sample risk-adjusted performance relative to the conventional 60/40 equity-bond benchmark. <br/>Using a Mean-Variance Optimization (MVO) framework enhanced by Ledoit-Wolf <br/>covariance shrinkage, the study evaluates seven distinct portfolio configurations across an <br/>empirical window spanning January 2016 to May 2026, covering multiple market regimes <br/>including the 2020 COVID-19 liquidity shock, the 2022 inflationary spike, and the 2024 <br/>institutional crypto adoption phase. <br/>The empirical design employs a dual-mode out-of-sample validation strategy: a fixed-weight <br/>stability test and a rolling realism test with systematic rebalancing. Results suggest that <br/>portfolios enriched with Bitcoin tended to achieve superior Sharpe and Sortino ratios relative <br/>to the baseline within this sample period and asset universe. The Master Portfolio (Case 7), <br/>combining equities, bonds, Bitcoin, Ethereum, Gold, and Silver, attains a static Sharpe Ratio <br/>of 1.62, a static Sortino Ratio of 3.23 and a Maximum Drawdown of only −6.40%. These <br/>findings are consistent with the primary hypothesis (H₁) that, within this empirical setting, <br/>Bitcoin can improve out-of-sample risk-adjusted performance, while also providing evidence <br/>consistent with the secondary hypothesis (H₂) that Bitcoin acts primarily as a portfolio <br/>diversifier rather than a consistent safe haven. All results should be interpreted as conditional <br/>on the sample period, the chosen asset universe, and the rebalancing assumptions. All asset <br/>price data was sourced from Investing.com and the risk-free rate from the FRED 3-Month <br/>Treasury Bill series (TB3MS).
The rapid expansion of the digital economy has exposed significant limitations in traditional economic frameworks, which struggle to explain phenomena such as algorithmic decisionmaking, data-driven value creation, and platform-based concentration. Existing approachesranging from production function extensions to platform models-remain fragmented and lack a unified micro-foundation. This paper proposes a behavior-centered framework to characterize economic forms and introduces the concept of economic morphology defined along four dimensions: agent structure, factor composition, behavioral pathways, and spatial distribution. Building on this framework, we define the Information Process Ratio (IPR) as a measurable indicator capturing the proportion of information-processing activities within economic behavior. Using IPR as a discriminant variable, we identify four major economic forms in human historyagricultural (IPR 10-20%), industrial (30-40%), service (50-60%), and digital (75-90%+). We show that the digital economy represents a distinct morphology, not a continuation of the industrial paradigm. Contemporary financial technology (FinTech) systems-high-frequency trading (HFT), decentralized finance (DeFi), and automated market makers (AMMs)represent extreme high-IPR regimes (95-99%), making them natural laboratories for testing the framework's predictions. We operationalize IPR using transaction-level proxies such as order-to-trade ratios (OTR), cancellation rates, and algorithmic trading share, enabling empirical application in financial markets. The framework generates testable implications linking IPR to transaction intensity, market concentration, returns to scale, algorithmic mediation, and high-frequency volatility. We further introduce the concept of IPR arbitrage, whereby economic activity flows toward higher-IPR systems, and propose a Financial Tension Index (FTI) to capture systemic strain in high-IPR environments. By shifting the analytical focus from agents to behaviors, this paper provides a unifying perspective for understanding the structural transformation of the digital economy and offers concrete implications for financial technology regulation, algorithmic market design, and systemic-risk monitoring.