This paper applies wavelet quantile correlation to research on the relationship among renewable energy stocks, investor sentiment, and the cryptocurrency market. The empirical results indicated that under extremely negative conditions, in both the short and medium run, renewable energy stocks and cryptocurrencies are negatively correlated, implying that during such periods, renewable energy stocks can be used as a safe haven for cryptocurrencies. The opposite happens when the market is average or booming. This indicates that investors tend to invest simultaneously in these two promising asset classes when the market performs well. Under varied market conditions, FGI correlates positively with cryptocurrency, demonstrating sentiment influences price patterns. Moreover, the correlation between FGI and renewable energy stocks further validates the relationship between cryptocurrencies and renewable energy stocks. These findings can be used to improve the prediction of market trends by investors using sentiment indices and to devise more effective portfolio diversification strategies that minimize risk amid an evolving market.
The rapid evolution of digital assets transforms cryptocurrencies into one of the most volatile and data-rich financial markets. Their nonlinear and unpredictable nature limits the effectiveness of traditional forecasting models, motivating the use of machine learning methods to identify hidden patterns and short-term price movements. This study compares the performance of Logistic Regression (LR), Random Forest (RF), XGBoost, Support Vector Classifier (SVC), K-Nearest Neighbors (KNNs), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) models in predicting the daily price directions of Bitcoin (BTC), Ethereum (ETH), and Ripple (XRP). Extensive data preprocessing and feature engineering are performed, integrating a broad set of technical indicators to enhance model generalization and capture temporal market dynamics. The results show that XGBoost achieves the highest classification accuracy of 55.9% for BTC and 53.8% for XRP, while LR provides the best result for Ethereum with an accuracy of 54.4%. In trading simulations, XGBoost achieves the strongest performance, generating a cumulative return of 141.4% with a Sharpe ratio of 1.78 for Bitcoin and 246.6% with a Sharpe ratio of 1.59 for Ripple, whereas LSTM delivers the best results for Ethereum with a 138.2% return and a Sharpe ratio of 1.05. Compared to recent studies, the proposed approach attains slightly higher accuracy, while demonstrating stronger robustness and profitability in practical backtesting. Overall, the findings confirm that through rigorous preprocessing machine learning-based strategies can effectively capture short-term price movements and outperform the conventional buy-and-hold benchmark, even under a simple rule-based trading framework.
Abstract This paper investigates the optimization of data sampling and target labeling techniques to enhance algorithmic trading strategies in cryptocurrency markets, focusing on Bitcoin (BTC) and Ethereum (ETH). Traditional data sampling methods, such as time bars, often fail to capture the nuances of the continuously active and highly volatile cryptocurrency market and force traders to wait for arbitrary points in time. To address this, we propose an alternative approach using information-driven sampling methods, including the CUSUM filter, range bars, volume bars, and dollar bars, and evaluate their performance using tick-level data from January 2018 to June 2023. Additionally, we introduce the Triple Barrier method for target labeling, which offers a solution tailored for algorithmic trading as opposed to the widely used next-bar prediction. We empirically assess the effectiveness of these data sampling and labeling methods to craft profitable trading strategies. The results demonstrate that the innovative combination of CUSUM-filtered data with Triple Barrier labeling outperforms traditional time bars and next-bar prediction, achieving consistently positive trading performance even after accounting for transaction costs. Moreover, our system enables making trading decisions at any point in time on the basis of market conditions, providing an advantage over traditional methods that rely on fixed time intervals. Furthermore, the paper contributes to the ongoing debate on the applicability of Transformer models to time series classification in the context of algorithmic trading by evaluating various Transformer architectures—including the vanilla Transformer encoder, FEDformer, and Autoformer—alongside other deep learning architectures and classical machine learning models, revealing insights into their relative performance.
This study contributes to the growing literature on the determinants of Bitcoin volatility by examining its relationship with financial stress. Building on prior research linking Bitcoin volatility to broader economic and financial uncertainty, we employ a combination of regression analysis, a GARCH-MIDAS framework, and a Vector Autoregression (VAR) model to evaluate both the static and dynamic effects of financial uncertainty on Bitcoin. Preliminary regression results indicate that financial stress measures significantly and negatively predict Bitcoin volatility. The GARCH-MIDAS model confirms these results, showing a strong negative impact of financial stress on the long-term component of volatility. VAR analysis further reveals that Bitcoin volatility decreases in response to shocks in financial stress indicators. These findings highlight Bitcoin’s sensitivity to systemic financial conditions and carry important implications for risk management among cryptocurrency traders, institutional investors, and financial regulators.
Blockchain technology is considered a transformative innovation, offering decentralized, secure, and transparent solutions to various industries, with cryptocurrencies being its most famous application. The volatility and non-linear behavior of cryptocurrency markets pose significant challenges for predicting their prices accurately. Predicting cryptocurrencies prices based on traditional statistical methods often fail to capture the market complex dynamics. Therefore, the recent developments in Artificial Intelligence, especially in deep learning and ensemble-based approaches have presented promising results. This study delivers a comprehensive literature review focusing on applying deep learning and ensemble deep learning algorithms in cryptocurrency time series price prediction. The main deep learning models such as Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Convolutional Neural Network (CNN), and Recurrent Neural Network (RNN) are examined with a variety of time intervals and cryptocurrency types. The findings present that deep learning models, especially when used in hybrid or ensemble configurations, have obtained promising results. This review highlights the efficacy and significant potential of ensemble deep learning and its capabilities in cryptocurrencies price trend forecasting offering valuable insights for investors and researchers.
Piotr Fiszeder, Witold Orzeszko, Radosław Pietrzyk, Grzegorz Dudek
This study advances the understanding of Bitcoin volatility forecasting by analysing an extensive set of 62 explanatory variables, including cryptocurrency market behaviour, Google search trends, financial indices, and economic indicators. We employ Bayesian Model Averaging (BMA), Least Absolute Shrinkage and Selection Operator (LASSO), and Random Forest (RF) methods to assess variable importance and forecast accuracy. Our research demonstrates that LASSO and RF models incorporating exogenous variables significantly improve both daily and weekly Bitcoin variance forecasts compared to models using only lagged Bitcoin volatilities. Key factors influencing Bitcoin volatility include lagged realised variances, trading volume, and Google search intensity. The study reveals that the impact of these variables on Bitcoin volatility is time-varying, reflecting its evolving relationship with broader economic indicators and market sentiment. Our findings contribute to the literature by providing a comprehensive analysis of Bitcoin volatility drivers, evaluating the effectiveness of variable transformations, and comparing the performance of advanced forecasting methods in handling the cryptocurrency's extreme volatility. These insights are valuable for researchers, investors, portfolio managers, and policymakers navigating the dynamic cryptocurrency market.
The increasing use of decentralized finance (DeFi) accelerates the demand for trustless, secure mechanisms for crosschain token exchange. This paper outlines a complete model for atomic token swaps based on the Hashed Timelock Contract (HTLC) scheme, allowing for intermediary-free token exchanges across disparate blockchain systems. The system makes use of the local blockchain simulation framework, Ganache, to design and test cross-chain interactions in a sandbox environment. To improve the decision-making capabilities for users, a real-time cryptocurrency price forecasting subsystem is added which utilizes machine learning models to analyze and predict the market and its volatility. Additionally, the system harnesses Generative AI capabilities through prompt engineering to tailor investment advice for individual users by analyzing the market, their preferred risk level, expected returns, and provide investment strategies aligned with users' preferences. Apart from sophisticated trading algorithms, the solution also offers a simple dashboard for market price monitoring and performs rapid token swaps at the user's command. Smart contracts are implemented using Solidity, token and price feeds are ports to Web3.js, predictive analytics is done in Python, while the frontend and backend are structured in Next.js alongside Node.js. System testing validates hypotheses on the provision of secure cross-chain swaps within one transaction without compromising.
Bitcoin’s extreme volatility continues to challenge accurate forecasting and risk management. Traditional econometric approaches struggle with the nonlinear and shifting dynamics of cryptocurrency markets, while deep learning models such as the Gated Recurrent Unit (GRU) often lack interpretability and adaptability to regime changes. To address these limitations, this study introduces a hybrid Gaussian Hidden Markov Model–Gated Recurrent Unit (HMM-GRU) framework for Bitcoin price forecasting. The HMM identifies latent market regimes from four years of daily closing prices and integrates these states as auxiliary features for the GRU network. Experimental results show that the hybrid model consistently surpasses the standalone GRU in predictive accuracy. Under the optimal configuration, HMM-GRU achieves a Mean Absolute Error (MAE) of 1,557.33 and a Mean Absolute Percentage Error (MAPE) of 1.42%, compared with 1,713.30 and 1.57% for GRU, representing an approximate 9% improvement in both absolute and relative error performance. The inclusion of regime-based features enables the model to better capture market transitions and mitigate overfitting to short-term noise. Beyond performance gains, the proposed approach enhances interpretability by linking forecasts to identifiable market regimes. These findings highlight the value of combining statistical regime detection with deep learning for volatile financial assets, providing practical insights for both investors and researchers in time-series forecasting.
Industry 4.0 and digital transformation have accelerated the emergence of virtual assets such as cryptocurrencies. Among them, Bitcoin, a virtual currency, has captured significant attention from both finance theorists and practitioners, achieving the highest market capitalization to date. The objective of this study is to examine the behavior and interrelationships between Bitcoin and several traditional financial assets within the framework of an international diversification strategy that combines conventional and crypto assets. In this context, Bitcoin is considered as a potential new asset class for portfolio diversification. To explore this relationship, we analyze the links between Bitcoin and a selection of major currencies—EUR, GBP, and JPY—as well as certain commodities. The study employs the Value at Risk (VaR) approach using three empirical methods, complemented by Conditional Value at Risk (CVaR) as a robustness measure, given its ability to capture tail risk more effectively than VaR. Using daily data from October 29, 2016, to October 23, 2020, the findings reveal that including Bitcoin in a diversified portfolio can significantly enhance risk–return characteristics. These results provide new insights for portfolio managers and investors seeking optimal diversification strategies in the context of digital finance.
Romulo de Moraes, Arthur G. Bubolz, Denner Ayres, Vinícius Teixeira Pinto · 5 authors
Este artigo apresenta uma arquitetura para loterias descentralizadas na rede Ethereum, baseada em contratos inteligentes, aleatoriedade verificável e otimização de armazenamento por meio de Merkle Tree. A proposta visa reduzir o custo médio das transações (gas fees) e aprimorar a escalabilidade onchain, comparando três abordagens distintas de armazenamento: array, mapping e Merkle Tree. Os resultados mostram que o consumo de gas evidencia uma vantagem expressiva da Merkle Tree, reduzindo em até três ordens de magnitude o custo total, o que confirma sua eficiência e potencial para aplicações descentralizadas de alta demanda.
We investigate the predictability of cryptocurrency returns using a comprehensive set of macroeconomic and cryptocurrency-specific factors and a set of 12 machine learning models. To enhance interpretability, we employ SHAP analysis to quantify the marginal contribution of each factor to model outputs. We further assess the economic value of predictive signals by constructing long-short and long-only portfolios. Empirically, tree-based methods, particularly random forests, deliver the highest predictive accuracy and outperform neural network and linear benchmarks, with predictability substantially stronger than that documented in equity markets. Across models, the market-to-realized-value ratio, new addresses, and active addresses consistently emerge as the most influential predictors, with higher values associated with higher expected returns. Portfolio results show that neural network-based strategies achieve the highest cumulative performance, indicating meaningful investment gains. Overall, our findings demonstrate the value of machine learning for return forecasting in the cryptocurrency market and provide practical insights for investors and financial analysts operating in highly volatile and evolving cryptocurrency environments.
Crypto analysts have to deal with a variety of challenges, with the most important area being the price volatility of cryptocurrencies. Due to uncertain market trends, many studies have been conducted on forecasting techniques, and some of these techniques have been integrated with advanced analytical tools, including machine learning (ML) techniques. Making reliable predictions of the speculative behavior of financial assets, especially in non-stationary and highly volatile environments such as the cryptocurrency market, is a challenging task. In this study, ML techniques are used to identify influential features that affect the prices of cryptocurrencies, especially for Bitcoins. In addition, multivariate control charts are utilized for signal detection, allowing for a structured approach to develop trading strategies for seasonal market conditions. Unlike other studies that do not take seasonality into serious consideration when analyzing market fluctuations, the proposed approach explicitly accounts for it. The developed strategy is tested across various market conditions, including the final days of each year from 2019 to 2024, and demonstrates strong and consistent performance in all cases. By systematically identifying key on-chain features and analyzing them by means of control charts, this study develops a structured approach to anomaly-based trading strategies in Bitcoins. These discoveries address an extensive discussion on automated trading systems, demonstrating that feature selection, technical indicators, market seasonality, and halving impacts are important components in hinting at successful cryptocurrency exchange strategies.
This study conducts a bibliometric review of Bitcoin research in the Business and Economics domains, using VOSviewer to visualize network structures and Bidirectional Encoder Representations from Transformers Topic (BERTopic) to derive semantically coherent topic clusters. The analysis identifies five major research themes: (1) Diversification, hedging, and safe-haven properties; (2) Market dynamics, efficiency, and investor behavior; (3) Bitcoin price and volatility prediction attempts; (4) Environmental impact of Bitcoin; and (5) Financial impact of Central Bank Digital Currency (CBDC). Based on these themes, the study recommends further investigation into the influence of Exchange-Traded Fund (ETF) approvals, regulatory frameworks, and institutional investor participation on Bitcoin’s safe-haven potential; the role of market dynamics and regulatory interventions; early detection of herding behavior and price bubbles; the integration of machine learning and deep-learning models for price prediction; the environmental costs associated with mining; and the evolving regulatory and implementation challenges of CBDCs. Overall, this review synthesizes existing scholarship and outlines future research directions for the rapidly evolving cryptocurrency ecosystem.
This study aims to examine the relationship between Bitcoin trading volume and key technical indicators using data-mining techniques to better understand how trading activity influences momentum and volatility in blockchain markets. The methodology involves analyzing a historical dataset of Bitcoin’s daily trading records from 2018 to 2023, which includes the Relative Strength Index (RSI), Moving Average Convergence Divergence (MACD), Simple and Exponential Moving Averages (SMA, EMA), and the Average True Range (ATR). Pearson correlation analysis was applied to identify linear associations between trading volume and these technical indicators. The results show significant positive correlations between trading volume and momentum or trend measures such as the 7-day RSI (r = 0.45, p < 0.05), SMA (r = 0.38, p < 0.05), EMA (r = 0.41, p < 0.05), and ATR (r = 0.48, p < 0.05), indicating that higher participation accompanies stronger market momentum and greater price variability. Conversely, the weak and non-significant correlation with MACD (r = –0.12, p = 0.15) suggests that volume has limited influence on lagging trend-reversal signals. The novelty of this study lies in integrating volume-based behavior into technical indicator analysis, extending the traditional volume–price–volatility framework to cryptocurrency markets and providing practical insights for momentum-driven trading strategies and volatility-aware risk management.
I Made Candra Girinata, Budi Styawan, Arwin Wahyu Saputra, M Aidil Arif · 5 authors
ABSTRAK Perkembangan aset kripto yang pesat, khususnya Ethereum, menuntut adanya model prediksi harga yang akurat untuk mendukung strategi investasi dan manajemen risiko. Penelitian ini bertujuan untuk menganalisis dan membandingkan kinerja dua algoritma machine learning ensemble, yaitu Random Forest (RF) dan XGBoost, dalam memprediksi harga harian Ethereum. Dataset historis ETH/USD sebanyak 3.423 observasi dari periode September 2016 hingga Juli 2025 diperoleh dari platform Bitfinex. Setelah melalui tahap pra-pemrosesan data dan rekayasa fitur temporal, dataset dibagi dengan rasio 80:20 untuk pelatihan dan pengujian. Model dievaluasi menggunakan metrik Root Mean Square Error (RMSE) dan Koefisien Determinasi (R²). Hasil eksperimen menunjukkan bahwa XGBoost secara signifikan mengungguli Random Forest, dengan nilai RMSE 134.63 dan R² 0.958. Sebagai perbandingan, Random Forest menghasilkan RMSE 208.45 dan R² 0.899. Temuan ini mengindikasikan bahwa mekanisme boosting pada XGBoost lebih efektif dalam menangkap kompleksitas dan volatilitas data pasar kripto. Kata kunci: Prediksi Harga, Ethereum, Machine Learning, XGBoost, Random Forest.
Blockchain-based financial ecosystems generate unprecedented volumes of multi-temporal data streams requiring sophisticated analytical frameworks that leverage both on-chain transaction patterns and off-chain market microstructure dynamics. This study presents an empirical evaluation of a two-class confidence-threshold framework for cryptocurrency direction prediction, systematically integrating macro momentum indicators with microstructure dynamics through unified feature engineering. Building on established selective classification principles, the framework separates directional prediction from execution decisions through confidence-based thresholds, enabling explicit optimization of precision–recall trade-offs for decentralized financial applications. Unlike traditional three-class approaches that simultaneously learn direction and execution timing, our framework uses post-hoc confidence thresholds to separate these decisions. This enables systematic optimization of the accuracy-coverage trade-off for blockchain-integrated trading systems. We conduct comprehensive experiments across 11 major cryptocurrency pairs representing diverse blockchain protocols, evaluating prediction horizons from 10 to 600 min, deadband thresholds from 2 to 20 basis points, and confidence levels of 0.6 and 0.8. The experimental design employs rigorous temporal validation with symbol-wise splitting to prevent data leakage while maintaining realistic conditions for blockchain-integrated trading systems. High confidence regimes achieve peak profits of 167.64 basis points per trade with directional accuracies of 82–95% on executed trades, suggesting potential applicability for automated decentralized finance (DeFi) protocols and smart contract-based trading strategies on similar liquid cryptocurrency pairs. The systematic parameter optimization reveals fundamental trade-offs between trading frequency and signal quality in blockchain financial ecosystems, with high confidence strategies reducing median coverage while substantially improving per-trade profitability suitable for gas-optimized on-chain execution.
Abstract This study evaluates the predictive accuracy of traditional time series (TS) models versus machine learning (ML) methods in forecasting realized volatility across major cryptocurrencies—Bitcoin (BTC), Ethereum (ETH), Litecoin (LTC), and Ripple (XRP). Employing high-frequency data, we analyze cross-cryptocurrency volatility dynamics through two complementary approaches: volatility forecasting and connectedness analysis. Our findings reveal three key insights: (i) TS models, particularly the heterogeneous autoregressive (HAR) model, exhibit superior predictive performance over their ML counterparts, with the long short-term memory (LSTM) model providing competitive yet inconsistent results due to overfitting and short-term volatility challenges; (ii) including lagged realized volatility of large-cap coins improves predictive accuracy for mid-cap coins, especially XRP, whereas forecasts for large-cap coins remain stable, indicating more resilient volatility patterns; and (iii) volatility connectedness analysis reveals substantial spillover effects, particularly pronounced during market turmoil, with large-cap assets (BTC and ETH) acting as primary volatility transmitters and mid-cap assets (XRP and LTC) serving as volatility receivers. These results contribute to the understanding of volatility forecasting and risk management in cryptocurrency markets, offering implications for investors and policymakers in managing market risk and interdependencies in digital asset portfolios.
Behavioral finance has emerged as a critical framework for understanding market dynamics beyond traditional rational agent models. This research presents a comprehensive multimodal approach to behavioral finance analysis, integrating market data, macroeconomic indicators, news sentiment, cryptocurrency metrics, Web3 analytics, GitHub development activity, and social sentiment to test five advanced hypotheses regarding behavioral pattern identification and market anomaly detection. The study employs an ultra-comprehensive data pipeline processing 30,400 samples across seven distinct data sources, generating 91 engineered features representing behavioral biases, investment patterns, and market psychology. Advanced machine learning techniques including Principal Component Analysis, t-Distributed Stochastic Neighbor Embedding, Variational Autoencoders, K-Means, Hierarchical Clustering, DBSCAN, Isolation Forest, One-Class SVM, and Elliptic Envelope are applied to identify behavioral structures and detect anomalies. Statistical validation through chi-square tests, ANOVA, Granger causality analysis, and lagged correlation studies demonstrates that three of five hypotheses (60%) achieve statistical significance at p < 0.05. Key findings reveal that behavioral structures exist and correspond to canonical biases (chi-square = 3406.780, p < 0.001), cluster assignments maintain moderate stability across market regimes (Jaccard similarity = 0.300), and sentiment and macroeconomic factors exhibit 65 significant causal relationships with behavioral patterns. However, multimodal data integration does not uniformly improve clustering quality (Silhouette score decrease of 0.116), and cluster-conditioned anomaly detection fails to outperform global methods (F1-score decrease of 0.017). These findings contribute to behavioral finance theory while providing practical applications for investment management, fraud detection, and regulatory compliance.
This paper examines factors that influence prices of most common five cryptocurrencies such as Bitcoin, Ethereum, Dash, Litecoin, and Monero over 2010-2018 using weekly data. The study employs ARDL technique and documents several findings. First, cryptomarket-related factors such as market beta, trading volume, and volatility appear to be significant determinant for all five cryptocurrencies both in short- and long-run. Second, attractiveness of cryptocurrencies also matters in terms of their price determination, but only in long-run. This indicates that formation (recognition) of the attractiveness of cryptocurrencies are subjected to time factor. In other words, it travels slowly within the market. Third, SP500 index seems to have weak positive long-run impact on Bitcoin, Ethereum, and Litcoin, while its sign turns to negative losing significance in short-run, except Bitcoin that generates an estimate of -0.20 at 10% significance level. Lastly, error-correction models for Bitcoin, Etherem, Dash, Litcoin, and Monero show that cointegrated series cannot drift too far apart, and converge to a long-run equilibrium at a speed of 23.68%, 12.76%, 10.20%, 22.91%, and 14.27% respectively.
Behavioral finance explores the psychological influences and cognitive biases that affect investor behavior and financial decision-making, including herding, the disposition effect, overconfidence, and others. Algorithmic trading is a method that uses computer programs to automatically execute buy and sell orders based on predefined mathematical models and trading strategies. With the continuous development of modern technology, the advent of the Web3 era, and the gradual evolution of artificial intelligence, algorithmic trading is becoming increasingly prevalent and garnering significant attention. While algorithmic trading is automated and may seem immune to human cognitive biases, the opposite is often true. This study aims to review the main findings of existing research from the perspective of the stock market, exploring the interactive relationship between behavioral finance and algorithmic trading and how cognitive biases such as herding and the disposition effect can influence algorithm performance. The results emphasize the importance of behavioral finance in both the research and practice of algorithmic trading, while also proposing the potential for using machine learning techniques to advance the field of behavioral finance. By integrating existing theories, this study contributes to a deeper understanding of the relationship between behavioral finance and algorithmic trading and offers new perspectives for its future development.
Vera Mita Nia, Hermanto Siregar, Roy Sembel, Nimmi Zulbainarni
This study explores how surprise shocks in Indonesia’s macroeconomic environment—specifically interest rates, inflation, and exchange rates—affect the returns and volatility of key financial assets, including gold, Bitcoin (BTC), stocks (JKSE), and government bonds. Utilizing the EGARCH(1,1) model, this research demonstrates that gold exhibits enduring resilience as a safe-haven during periods of rising inflation and interest rate fluctuations. In contrast, Bitcoin is marked by pronounced speculative dynamics, showing persistent, asymmetric, and extreme volatility, yet delivering attractive gains when market conditions are strong. The findings indicate that stocks and bonds are particularly susceptible to changes in macroeconomic variables, thereby illustrating the vulnerabilities typical of emerging markets. Through portfolio optimization employing the Mean-Variance approach, gold dominates the optimal asset allocation, while Bitcoin provides notable diversification benefits. The results of backtesting using the Kupiec and Basel Traffic Light procedures confirm that GARCH-family risk estimations are robust and meet international regulatory standards. Furthermore, analysis of the Sharpe ratio and cumulative returns reveals that Mean-Variance portfolios consistently outperform equally weighted alternatives by delivering higher risk-adjusted returns and lower overall volatility. By integrating advanced econometric methods with real-world macroeconomic shocks in an Indonesian context, this research offers practical insights for both investors and policymakers addressing asset allocation under uncertainty, while laying the groundwork for future work involving broader asset universes and sophisticated modeling techniques.
Large Language Models (LLMs) are increasingly deployed in high-stakes financial domains, yet they suffer from specific, reproducible hallucinations when performing arithmetic operations. Current mitigation strategies often treat the model as a black box. In this work, we propose a mechanistic approach to intrinsic hallucination detection. By applying Causal Tracing to the GPT-2 XL architecture on the ConvFinQA benchmark, we identify a dual-stage mechanism for arithmetic reasoning: a distributed computational scratchpad in middle layers (L12-L30) and a decisive aggregation circuit in late layers (specifically Layer 46). We verify this mechanism via an ablation study, demonstrating that suppressing Layer 46 reduces the model's confidence in hallucinatory outputs by 81.8%. Furthermore, we demonstrate that a linear probe trained on this layer generalizes to unseen financial topics with 98% accuracy, suggesting a universal geometry of arithmetic deception.
S Thillainayagi, Paolo Pavan, S Shashank, L V Preetham · 5 authors
Bitcoin is known for its high volatility and speculative trading behavior.Predicting Bitcoin prices is valuable for investors, traders, and financial analysts.The study uses historical price data, technical indicators, and/or sentiment analysis.Machine learning and statistical models like ARIMA, Linear Regression, and LSTM are applied.Deep learning models, especially LSTM, show better accuracy in capturing time-series patterns