Bitcoin is known for its high volatility, which makes it challenging to accurately predict future prices. In this study, we aim to forecast Bitcoin prices for a month by incorporating exogenous variables, specifically the interest rate and recession probability. Our primary objective is to explore whether these variables have a positive impact on the prediction of Bitcoin prices. We used two popular time series forecasting models: Long Short-Term Memory (LSTM) and Facebook Prophet. Our approach involves exploring the impact of these exogenous variables on the performance of the models and comparing their results through plots and cross-validation. We trained the models using historical Bitcoin price data along with exogenous variables and evaluated their performance on a test dataset. Our results indicate that LSTM outperforms Facebook Prophet in terms of Bitcoin price prediction accuracy. This is because, while Facebook Prophet is optimized for statistical forecasting modeling, LSTM has the capability to learn intricate patterns and relationships given the right architecture with sufficient neurons. Importantly, we demonstrate that incorporating interest rates and recession probabilities significantly enhances the predictive capability of our models. Our findings suggest that changes in interest rates and recession probabilities have an impact on Bitcoin prices, and our models perform better when equipped with this valuable information.
David Ardia, David Ardia, Keven Bluteau, Keven Bluteau
We study the relation between the promotion of a cryptocurrency on Twitter and its return dynamics around pump-and-dump events. By analyzing abnormal returns, trading volume, and tweet activity, we uncover that Twitter effectively garners attention for pump-and-dump schemes, leading to notable effects on abnormal returns before the event. Our results indicate that investors relying on Twitter information exhibit delayed selling behavior during the post-dump phase, resulting in significant losses compared to other participants. We also find that, while tweets directly promoting pump schemes align with anticipated market phases, a noteworthy portion of indirect, non-pump-aware tweets significantly influence market movements pre-event.
The rapid evolution of cryptocurrency markets and the increasing complexity of trading strategies necessitate a comprehensive understanding of price-prediction models and their direct impact on trading efficacy. While extensive research has been conducted separately on price prediction methods and trading strategies, there remains a significant gap in studies explicitly correlating precise price forecasts with successful trading outcomes. This review paper addresses this gap by critically examining the role of accurate cryptocurrency price predictions in enhancing trading strategies. We conducted a systematic review of sufficient scholarly articles and web resources, focusing on the methodologies and effectiveness of various predictive models and their integration into cryptocurrency trading strategies. Our selection criteria ensured the inclusion of papers that demonstrate methodological rigor, relevance, and recent contributions to the field, spanning from economic theories and statistical models to advanced machine learning techniques. The findings reveal that precise price predictions significantly contribute to the development of adaptive and risk-managed trading strategies, which are crucial in the highly volatile cryptocurrency market. The review also identifies current challenges and proposes directions for future research, emphasizing the need for interdisciplinary approaches and ethical considerations in predictive modeling. This synthesis aims to bridge the existing research gap and guide future studies, thereby fostering more sophisticated and profitable trading strategies in the cryptocurrency domain.
Cryptocurrencies, recognized by their extreme volatility due to dependency on multiple direct and indirect factors, offer a significant challenge regarding precise price forecasting. This uncertainty has led to investment hesitation within the digital currency market. Previous research attempts have presented methodologies for price forecasting and trend prediction in cryptocurrencies. However, these forecasts have typically suffered from increased error rates, leaving the opportunity for improvement in this field. Furthermore, the influence of sentiment-based factors could compromise the reliability of price predictions. In this research, we have proposed a machine learning-driven framework that provides precise cryptocurrency price projections and adds an alert mechanism to guide investors. Our fundamental analyzer, Bi-LSTM and GRU hybrid model use historical data of digital currencies to train and reliably anticipate future values. Complementing this, a sentiment analyzer, utilizing a BERT and VADER hybrid model, analyzes sentiments to assess the forecast price as trustworthy or uncertain. Besides assisting investor decision-making, this technique also helps risk management in the dynamic realm of cryptocurrency. Our suggested approach delivers highly precise price predictions with dramatically decreased error rates compared to prior competitive studies. The proposed Bi-LSTM-GRU-BERT-VADER (BLGBV) model is tested for three cryptocurrencies, namely BTC, ETH, and Dogecoin and reports an average root mean square error (RMSE) of 0.0241%, 0.0645%, and 0.0978%, respectively.
Krzysztof Gogol, Robin Fritsch, Malte Schlosser, Johnnatan Messias · 6 authors
This paper studies liquid staking tokens (LSTs) on automated market makers (AMMs), both theoretically and empirically. LSTs are tokenized representations of staked assets on proof-of-stake blockchains. First, we model LST-liquidity on AMMs theoretically, categorizing suitable AMM types for LST liquidity and deriving formulas for the necessary returns from trading fees to adequately compensate liquidity providers under the particular price trajectories of LSTs. For the latter, two relevant metrics are considered: (1) losses compared to holding the liquidity outside the AMM (loss-versus-holding, or "impermanent loss"), and (2) the relative profitability compared to fully staking the capital (loss-versus-staking) which is specifically tailored to the case of LST-liquidity. Next, we empirically measure these metrics for Ethereum LSTs across the most relevant AMM pools. We find that, while trading fees often compensate for impermanent loss, fully staking is more profitable for many pools, raising questions about the sustainability of the current LST liquidity allocation to AMMs.
In the Bitcoin trading landscape, predicting price movements is paramount. Our study focuses on identifying the key factors influencing these price fluctuations. Utilizing the Pearson correlation method, we extract essential data points from a comprehensive set of 14 data features. We consider historical Bitcoin prices, representing past market behavior; trading volumes, which highlight the level of trading activity; network metrics that provide insights into Bitcoin’s blockchain operations; and social indicators: analyzed sentiments from Twitter, tracked Bitcoin-related search trends on Google and on Twitter. These social indicators give us a more nuanced understanding of the digital community’s sentiment and interest levels. With this curated data, we forge ahead in developing a predictive model using Deep Q-Network (DQN). A defining aspect of our model is its innovative reward function, tailored for enhancing predicting Bitcoin price direction, distinguished by its multi-faceted reward function. This function is a blend of several critical factors: it rewards prediction accuracy, incorporates confidence scaling, applies an escalating penalty for consecutive incorrect predictions, and includes a time-based discounting to prioritize recent market trends. This composite approach ensures that the model’s performance is not only precise in its immediate predictions but also adaptable and responsive to the evolving patterns of the cryptocurrency market. Notably, in our tests, our model achieved an impressive F1-score of 95%, offering substantial promise for traders and investors.
In the realm of cryptocurrency, the prediction of Bitcoin prices has garnered substantial attention due to its potential impact on financial markets and investment strategies. This paper propose a comparative study on hybrid machine learning algorithms and leverage on enhancing model interpretability. Specifically, linear regression(OLS, LASSO), long-short term memory(LSTM), decision tree regressors are introduced. Through the grounded experiments, we observe linear regressor achieves the best performance among candidate models. For the interpretability, we carry out a systematic overview on the preprocessing techniques of time-series statistics, including decomposition, auto-correlational function, exponential triple forecasting, which aim to excavate latent relations and complex patterns appeared in the financial time-series forecasting. We believe this work may derive more attention and inspire more researches in the realm of time-series analysis and its realistic applications.
Cryptocurrencies are changing how we view and interact with traditional currencies, and they have become a disruptive force in the financial industry. Accurate price prediction is becoming more and more important as the bitcoin industry grows in size and complexity. This paper provides a thorough examination of deep learning models used in bitcoin price prediction. We explore the dynamic and unpredictable character of the cryptocurrency market, where price swings can happen quickly and without warning. To comprehend the present state of the art in this domain and pinpoint the shortcomings of the deep learning models in use today, we examine the body of existing literature. The data collecting and preprocessing methods used to get the bitcoin market data ready for modeling are described in the methodology section. Numerous deep learning models—Recurrent Neural Networks among them, Convolutional neural networks (CNNs) and Long Short-Term Memory (LSTM) networks are investigated. We go over hyperparameter tweaking, model training, and the assessment metrics that are used to gauge the performance of the model. We offer a thorough case study that focuses on forecasting the price of a particular cryptocurrency, like Bitcoin, in order to offer empirical insights. Our results provide light on the difficulties and possibilities involved in this project, emphasizing the need for creative solutions to address the market
Amaç–Bitcoin başta olmak üzere kripto varlık fiyatlarında meydana gelen hızlı değişimler gerek finansal yatırımcı gerekse medya tarafından ilgi görmektedir. Buna bağlı olarak kar elde etmek amacı başta olmak üzere pek çok farklı motivasyonla birçok araştırmacı ve finansal aktör, Bitcoin fiyatını etkileyen çeşitli faktörleri belirlemeye çalışmaktadır. Bitcoin fiyat hareketlerini etkilediği düşünülen Fed faiz oranı, altın ve Bitcoin’in farklı fiyat göstergeleri gibi öznitelikler üzerine detaylandırılan bir inceleme yürütülmektedir. Bu bağlamda fiyatları tahmin etmek için kullanılan çeşitli makine öğrenme algoritmaları üzerinde sistematik bir analiz yapılmaktadır.Yöntem –Farklı dört makine öğrenme modeli kullanılmış olup farklı tahmin hata oranları elde edilmiş ve her birinin çalışmada kullanılabileceği görülmüştür. Bulgular –Bitcoin veri seti için önerilen en iyi tahmin performansının sırasıyla Tesadüfi Ağaç (RF) %96,38, Karar Ağacı (DT) %96,28, Doğrusal Regresyon 95,06 ve Stokastik Gradient Descent(SGD) Doğrusal Regresyon %93,91 şeklinde olduğunu göstermektedir. Bitcoin fiyat değişimlerininFed faiz oranı ve altından ziyade kendi fiyat değişimlerinden daha yüksek oranda etkilendiği diğer sonuçlar arasında yer almaktadır. Tartışma –Tahmin modellemesinde en iyi sonuçları veren iki algoritmaya bakıldığında, gün içi en yüksek fiyatın son derece etkili olduğu söylenebilmektedir. En düşük fiyat ise ikinci derece en etkili özniteliktir. Söz konusu sonuç, Bitcoin’in en çok kendi fiyat dalgalanmalarından etkilendiğini göstermektedir.
This paper investigates an ensemble convolutional and recurrent neural network architecture for cryptocurrency price forecasting. The inherent volatility and noise in cryptocurrency time series pose considerable modeling challenges. The proposed ensemble model integrates convolutional neural networks (CNN) and gated recurrent units (GRU) to jointly discern spatial patterns and temporal dynamics. The model is trained on an extensive dataset comprising daily historical prices for major cryptocurrencies spanning January 2015 to October 2023. The time series data is structured into rolling input sequences of historical prices and target outputs as future price values. Comprehensive hyperparameter tuning is conducted to optimize model performance. Rigorous validation on held-out test data enables analysis of multi-step prediction accuracy. Results demonstrate that the ensemble CNN-GRU model achieves high forecasting proficiency. Evaluation metrics including Root Mean Squared Error quantify the model's efficacy in learning the nuanced volatility signatures of cryptocurrencies. Additionally, the high R-squared scores attained, including 0.99 for Bitcoin and Ethereum and 0.98 for Ripple, underscore the model's exceptional capacity to explain cryptocurrency price fluctuations. This substantiates the model's utility for generating actionable insights for investors and analysts in the cryptocurrency domain.
This paper describes an architecture for predicting the price of cryptocurrencies for the next seven days using the Adaptive Network Based Fuzzy Inference System (ANFIS). Historical data of cryptocurrencies and indexes that are considered are Bitcoin (BTC), Ethereum (ETH), Bitcoin Dominance (BTC.D), and Ethereum Dominance (ETH.D) in a daily timeframe. The methods used to teach the data are hybrid and backpropagation algorithms, as well as grid partition, subtractive clustering, and Fuzzy C-means clustering (FCM) algorithms, which are used in data clustering. The architectural performance designed in this paper has been compared with different inputs and neural network models in terms of statistical evaluation criteria. Finally, the proposed method can predict the price of digital currencies in a short time.
As a consequence of rising geo-economic issues, global currency values have declined during the last two years, stock markets have performed poorly, and investors have lost money. Consequently, there is a renewed interest in digital currencies. Cryptocurrency is a fresh kind of asset that has evolved as a result of fintech innovations, and it has provided a major research opportunity. Due to price fluctuation and dynamism, anticipating the price of cryptocurrencies is difficult. There are hundreds of cryptocurrencies in circulation around the world and the demand to use a prediction system for price forecasting has increased manifold. Hence, many developers have proposed machine learning algorithms for price forecasting. Machine learning is fast evolving, with several theoretical advances and applications in a variety of domains. This study proposes the use of three supervised machine learning methods, namely linear regression, support vector machine, and decision tree, to estimate the price of four prominent cryptocurrencies: Bitcoin, Ethereum, Dogecoin, and Bitcoin Cash. The purpose of this study is to compute and compare the precision of all three techniques over all four datasets.
Forecasting the value of Ethereum (ETH) or any other cryptocurrency is a formidable undertaking owing to the inherent volatility and speculative characteristics shown by these digital assets. Nevertheless, it is possible to create price forecasts by using machine learning techniques, namely Recurrent Neural Networks (RNNs), which are capable of capturing temporal relationships within the data. The challenge of forecasting the price of Ethereum (ETH) or any cryptocurrency is a multifaceted endeavour that encompasses aspects of finance, economics, and data science. The practise of technical analysis is the examination of past price charts, patterns, and technical indicators in order to make forecasts about future price fluctuations. The underlying assumption is that previous pricing patterns had the capacity to provide valuable insights into future developments. Nevertheless, it is essential to acknowledge that the effectiveness of technical analysis within the realm of cryptocurrency trading is a subject that engenders much scholarly discourse. The primary objective of this research is to examine the utilisation of Ethereum cryptocurrency and forecast its behaviour via the use of machine learning methodologies, namely Recurrent Neural Networks. The suggested approach demonstrates a high level of accuracy, reaching 95 percent. This significant level of precision will be beneficial for future academics working on this technology.
This paper illustrates the working process of predicting the Bitcoin price applying ARIMA, SARIMA and linear regression. Since more and more machine learning models were developed and tested in the financial field, these three models are selected to examine their reliabilities. In this study, three methodologies have been used for the Bitcoin predictions under the data set of Bitcoin historical prices. With the help of python notebook, order (1, 1, 1) and seasonal order (0, 1, 1, 12) were applied to the predictions in ARIMA and SARIMA respectively. In terms of linear regression, this paper used two independent variables including historical data and trading volume to predict the Bitcoin prices. It was discovered that the predictive graph for these three methodologies can match the actual value well, and linear regression performs the best. Considering the rapid development of machine learning methods, adopting alternative methods deserve in-depth investigations.
Ahmed Bouteska, Mohammad Zoynul Abedin, Petr Hájek, Kunpeng Yuan
Cryptocurrency price forecasting is attracting considerable interest due to its crucial decision support role in investment strategies. Large fluctuations in non-stationary cryptocurrency prices motivate the urgent need for accurate forecasting models. The lack of seasonal effects and the need to meet a number of unrealistic requirements make it difficult to make accurate forecasts using traditional statistical methods, leaving machine learning, particularly ensemble and deep learning, as the best technology in the area of cryptocurrency price forecasting. This is the first work to provide a comprehensive comparative analysis of ensemble learning and deep learning forecasting models, examining their relative performance on various cryptocurrencies (Bitcoin, Ethereum, Ripple, and Litecoin) and exploring their potential trading applications. The results of this study reveal that gated recurrent unit, simple recurrent neural network, and LightGBM methods outperform other machine learning methods, as well as the naive buy-and-hold and random walk strategies. This can effectively guide investors in the cryptocurrency markets.
The study aims to predict the close prices of four different cryptocurrencies (Bitcoin, Ethere-um, Dogecoin, and Cardano) using machine learning techniques and determine which of these cryptocurrencies is suitable for investment. To achieve this goal, we used two popular gradi-ent boosting algorithms: Extreme Gradient Boosting (XGBoost) and Light Gradient-Boosting Machine (LightGBM). Prediction accuracy of the trained model is evaluated by Mean Abso-lute Error (MAE) generated by the methodology of Cross-Validation. Our results show that both XGBoost and LightGBM can effectively predict the close prices of the four cryptocur-rencies, with LightGBM achieving slightly better performance in terms of prediction accura-cy. Based on our analysis, we were able to identify which cryptocurrencies were suitable for investing and provide recommendations for potential investors. Overall, our study highlights the potential of machine learning techniques in predicting cryptocurrency close prices and identifying suitable investment opportunities.
<span>The majority of research on predicting the price of Bitcoin employs technical methods to enhance long short-term memory models' effectiveness. Although some studies employ different machine learning techniques, such as economic or technical indicators, their precision is inadequate. Thus, this research aims to introduce a model that predicts the price of Bitcoin by utilizing the long short-term memory (LSTM) technique and incorporating gold's economic and technical data as features. The research collected gold and Bitcoin price data from FinanceDataReader for around seven years, from January 1, 2016, to January 22, 2023, consisting of six categories: date, open, high, low, close, volume, and change (based on dollars). The normalized closing price data was trained for 50 epochs, resulting in the loss value reaching close to zero. The model's accuracy was measured by mean squared error, resulting in a score of 0.0004. This study's importance is two-fold: firstly, it can provide cryptocurrency-related businesses with more accurate predictions and improved risk management indicators. Secondly, incorporating economic metrics can address the limitations of overfitting and a single model's poor performance.</span>
As cryptocurrencies become widely accepted due to technical improvements, reliable approaches to capture their future price movements of them become critical. This study mainly combines weighted sentiment analysis results from social media-related comments and financial news headlines with a stacked LSTM model to predict second-day Bitcoin price evolution. This study also compared our results and the results produced by MLP, RF, and SVM after feeding the sentiment analysis results.
Peter T. Yamak, Yujian Li, Ting Zhang, Kyefondeme C. Dakurah
This research proposes a novel approach for forecasting cryptocurrency prices, specifically Bitcoin which dominates the market. Accurately predicting cryptocurrency values is challenging due to their highly volatile nature. The proposed hybrid model uses ResNet Convolutional Neural Network to encode Bitcoin price time series data into discriminative representations. These representations capture long-range dependencies using XGBoost regression. Additionally, wavelet denoising is applied to filter noise from the price data. The combined ResNet-XGBoost-Wavelet model achieves satisfactory results for Bitcoin price forecasting and has practical applications for developing quantitative trading strategies. While incorporating sentiment analysis and additional influencing factors could further improve predictions, this work presents a competitive approach for minimizing investment risks and maximizing profits in the complex domain of cryptocurrency markets.
Guzalxon Belalova, Shakhida Gaybullaevna Mannanova, Botirjon Karimov
Over the past few decades, recurrent neural networks, particularly the Long Short-Term Memory (LSTM) architecture, have undergone several refinements. These networks have emerged as the go-to models for numerous machine learning challenges, especially those involving sequential data. One such application is the prediction of Bitcoin prices, a cryptocurrency that stands at the forefront of blockchain technology. This paper delves into the intricacies of forecasting Bitcoin prices using a suite of models, with a keen emphasis on the LSTM architecture, renowned for its prowess in handling tasks with long-term dependencies. Our exploration encompasses traditional time series models like ARIMA, neural network variants such as ANN and Transformer-based models, and even hybrid combinations. Specifically, our LSTM model, augmented with peephole connections, demonstrates its capability to learn and predict Bitcoin price fluctuations. We source our data from the Bitcoin Price Index and aim to gauge the accuracy with which these models can predict Bitcoin's price trajectory. Furthermore, our experiments involve the deployment of an "adam"-optimized Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) network, revealing insights into their predictive performances.
Alexandru Costin Baroiu, Vlad Dıaconıța, Simona‐Vasilica Oprea
Cryptocurrencies have emerged as a popular investment vehicle, prompting numerous efforts to predict market trends and identify metrics that signal periods of volatility. One promising approach involves leveraging on-chain data, which is unique to cryptocurrencies. On-chain data, extracted directly from the blockchain, provides valuable information, such as the hash rate, total transactions, or the total number of addresses that hold a specified amount of cryptocurrency. Some studies have also explored the relationship between social media sentiment and Bitcoin, using data from platforms such as Twitter and Google Trends. However, the quality of Twitter sentiment analysis has been lackluster due to suboptimal extraction techniques. This research proposes a novel approach that combines a superior sentiment analysis technique with various on-chain metrics to improve predictions using a deep learning architecture based on long-short term memory (LSTM). The proposed model predicts outcomes for multiple time horizons, ranging from one day to 14 days, and outperforms the Martingale (random walk) approach by over 9%, as measured by the mean absolute percentage error metric, as well as recent results reported in literature. To the best of our knowledge, this study may be among the first to employ this combination of techniques to improve cryptocurrency market prediction.
This paper focuses on the prediction of Bitcoin prices and returns based on the Long Short Term Memory (LSTM) neural network model, to better consider the impact of time factors. Since Bitcoin has long dominated the digital currency trading market, many researchers have completed many Bitcoin prediction results, including the screening of optimal features, comparison of prediction models and classification of prediction problems. Based on previous work, this article adds a Bitcoin revenue forecast section, presenting the results in the form of charts and data to provide more intuitive trends and more accurate performance. This paper uses LSTM as the experimental model, and uses the Bitcoin transaction history data set with timestamps as the original input. After a specific normalization method, the original model is trained, and then the subsequent transaction data is predicted. Compare it with the real value in the data set to get the final experimental results show that in this prediction problem, the performance of LSTM is slightly better than Autoregressive Integrated Moving Average (ARIMA) and eXtreme Gradient Boosting (XGBoost); on the other hand, compared with price prediction based on real values for prediction, the prediction fluctuations of return are more obvious and more realistic, providing better reference value.