M. Geetha Jenifel, R. Anita Jasmine, D Umanandhini
Cryptocurrency markets, exemplified by the notable volatility in Bitcoin prices, have become pivotal arenas for financial exploration and investment. In response to this, our research undertakes a comprehensive comparison of diverse machine-learning models to predict Bitcoin prices. The models scrutinized include Linear Regression, Ridge Regression, Decision Tree, Random Forest, Support Vector Machine (SVM), K-nearest neighbors (KNN), and Neural Networks. This study evaluates and contrasts these models based on performance metrics such as Mean Absolute Error (MAE), Mean Squared Error (MSE), and R -squared. This work encompasses the collection and preprocessing of historical Bitcoin price data, the engineering of pertinent features, and the division of the dataset into training and test sets. Each machine learning model is meticulously trained on the training set, allowing for the tuning of hyperparameters to optimize predictive capabilities. Subsequently, the models’ performance is systematically assessed on the test set, providing valuable insights into their accuracy and reliability. The scope of this research is delimited by a specific timeframe, focusing on historical Bitcoin price data up to 15/11/2023. By addressing these objectives, this research aspires to guide investors, researchers, and analysts in navigating the intricate landscape of cryptocurrency investment decisions. The findings contribute to an enhanced understanding of the nuanced strengths and limitations inherent in different machine learning models when applied to the volatile context of Bitcoin price prediction. Ultimately, this research aims to facilitate more informed and strategic decision-making processes in the cryptocurrency market.
The proposed work explores the significance of Bitcoin in today’s financial landscape and its role as a decentralized store of value and hedge against economic uncertainty. The diverse forecasts for Bitcoin prices are proposed and the importance of accurate prediction models. Specifically, it emphasizes the effectiveness of Long Short-Term Memory (LSTM) and Bidirectional LSTM (Bi-LSTM) models in capturing the complex dynamics of cryptocurrency markets, offering insights for traders, investors, and researchers. The growing importance of Deep Neural Networks (DNNs) is highlighted by analyzing historical market data and forecasting future price movements. The proposed work concludes by underscoring the evolution of Bitcoin price prediction methodologies from traditional models like ARIMA to advanced techniques like Bi-LSTM, and Auto-regressive EncoderDecoder Transformer, enhancing financial security and decision-making in the cryptocurrency market. The Bi-LSTM worked by providing a 0.9832 R2 score.
Cryptocurrencies, such as Bitcoin, Binance, Ethereum, FTX, and XRP, are decentralized digital assets known for their volatile nature and potential as investment instruments. Accurate price prediction is crucial for informed investment decisions. This study explores the feasibility of various modeling techniques on diverse data structures and features for predicting the prices of these cryptocurrencies. We utilize daily and high-frequency price data to classify and predict prices using deep learning and machine learning techniques, including LSTM, Bi-LSTM, GRU, linear regression, and SGD regression. Our findings indicate that daily price projections achieve an accuracy of 0.99, outperforming more complex deep learning and machine learning models. Compared to benchmark results, our approach demonstrates superior performance, with the highest scores achieved by the applied statistical methods and advanced algorithms. This research highlights the effectiveness of deep learning and machine learning models in cryptocurrency price forecasting, offering a foundation for further exploration in the field and emphasizing the significance of sample size in predictive modeling.
R. C. Bose, Jeevan Nagarkar, Sushant Malik, Nisha Bharti
Predictability of the various financial instruments can lead to more trust and investment. The study examines the long-term and causal relationship between various Nifty indices and the Ethereum cryptocurrency. This study considers the data from April 2015 to December 2022 in two phases, pre-covid and post-covid. Johansen’s cointegration test was used to determine if the vectors in the data set are cointegrated, using the Max-Eigen and Trace tests for evaluation. The Granger causality test was also used to explore the short-term causal relationship between Ethereum and the five Nifty indices. The study found that post-pandemic daily returns of stock market indices have developed a significant cointegration with the cryptocurrency over time. The Granger causality test results showed bi-directional relationships of Nifty 50, Nifty 200 and Nifty Next 50 with Ethereum and a unidirectional relationship between Nifty Auto and Ethereum. The non-linear results reveal a one-way relationship pre-covid and a bi-directional relationship post-covid except for Nifty Banks. Johansen’s cointegration test, both in the pre-and post-covid-era, indicated that these indices had a substantial long-term cointegration with cryptocurrencies. This study also offers guidance to investors in making long-term investment decisions and to regulatory authorities. This implied that the investing decisions resulted in developing a causal relationship between the equity market and cryptocurrencies, which seemed very unlikely before 2020. This indicates that a new and young investor also considered cryptocurrencies a viable alternate investment option compared to traditional options such as fixed deposits, gold, and other fixed-income instruments.
Gurpreet Singh, Ashok Kumar Sahoo, G. S. Yadav, Ajay Gairola · 6 authors
Crypto currencies represent a digital form of currency, entirely reliant on electronic transactions and lacking a physical counterpart in the form of traditional banknotes. Unlike fiat currencies, they operate in a decentralized manner, free from third-party intervention, enabling users to access services directly. Nevertheless, the volatility in cryptocurrency prices significantly impacts international relations and trade, contributing to economic inequalities on a global scale. This research concentrates on Bitcoin price prediction, a highly popular cryptocurrency widely accepted by various stakeholders, including investors, researchers, traders, and policymakers. Therefore, in this research, it is proposed to compare the performance of two machine learning models (a RNN model and a LSTM model) for bitcoin price prediction. Open and close pricing are the main requirements for implementation of the model. In addition, the research compares the accuracy values of both models for closed prices, contributing to Sustainable Cities and Communities. The results prove that LSTM model is a better choice than RNN model with a very low error of 0.196%.
A capacity of foreseeing price fluctuations in bitcoin with exceptionally precise is very worthwhile to investigators and funding sources. However, as the cryptocurrency market is nonlinear, it can be challenging to determine the distinctive features of time-series data, which renders it challenging to forecast accurate price estimations. Massive oscillations in non-stationary cryptocurrency values underscore the pressing necessity to precise forecasting models. The most effective methodology for cryptocurrency price forecasting is machine learning, foremost ensemble and deep learning. Traditional statistical methods are difficult to execute accurately because to the lack of seasonal variations and the need to meet a number of naive requirements. The suggested methodology builds upon the random walk theory, commonly utilized in financial markets to model stock values. To simulate market volatility, this methodology utilizes randomization into the observed feature activations of neural networks at a layer-wise level. Moreover, a mechanism to assess the market’s reaction pattern is incorporated into the prediction model. Training was conducted on ARIMA and LSTM, short for Long Short-Term Memory models using Ripple, Ethereum, and Tron as illustrative examples.
The recurring neural network is a deep learning algorithm that is commonly used to develop prediction systems. There are many variants of RNN such as RNN itself, long-short-term memory (LSTM), and gated recurring unit, so it is frequently debatable which algorithm from the RNN family has the most optimal efficiency and computation time. When developing a prediction system, sequential or time series data is required so that an accurate prediction can be made. Sequential or time series data involve data arranged in a time sequence, such as weather data, financial data, carbon emission data, and traffic data recorded over time. This research will be carried out by predicting the three RNN models against historical Bitcoin value data. The research method used is Experimental Design by comparing the performance between the three models on bitcoin value time series data, testing is done by involving hyperparameters such as Tanh, Sigmoid, and ReLU activation functions, batch size, and epochs. The aim of this research is to find out which RNN model can produce the most optimal performance and find out what performance measures can be used to evaluate and compare the performance between the three models. The results of the study show that LSTM is the most effective model with RMSE 0.012441 and MSE 0.000155 but inefficient because it takes 3 minutes 24 seconds to run the computation; in the meantime, the Tanh activation function gives the most optimal prediction than Sigmoid and RelU and therefore should be the main candidate to be used with RNN models when predicting Bitcoin prices.
This article presents a novel approach to cryptocurrency price forecasting, leveraging advanced machine-learning techniques.By comparing traditional autoregressive models with recurrent neural network approaches, the study aims to evaluate the forecasting accuracy of Autoregressive Integrated Moving Average (ARIMA), Seasonal ARIMA (SARIMA), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) models across various cryptocurrencies, including Bitcoin, Ethereum, Dogecoin, Polygon, and Toncoin.The data for this empirical study was sourced from historical prices of these specific cryptocurrencies, as recorded on the CoinMarketCap platform, covering January 2022 to April 2024.The methodology employed involves rigorous statistical and neural network modelling where each model's parameters were meticulously optimized for the specific characteristics of each cryptocurrency's price data.Performance metrics such as Mean Squared Error (MSE), Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE) were used to assess the precision of each model.The main results indicate that LSTM and GRU models, leveraging deep learning techniques, generally outperformed the traditional ARIMA and SARIMA models regarding error metrics.This demonstrates a higher efficacy of neural networks in handling the non-linear complexities and volatile nature of cryptocurrency price movements.This study contributes to the ongoing discourse in financial technology by elucidating the practical implications of using advanced machine-learning techniques for economic forecasting.Importantly, it provides valuable insights that can directly inform and enhance the decision-making processes of investors and traders in digital assets.
Cryptocurrencies are digital assets that have attracted a lot of investment and attention. It is challenging and essential for investors and traders to predict their stock price movements. Making accurate predictions about cryptocurrency prices is crucial for avoiding losses and gaining profits. Our research proposes a novel method for predicting the stock closed prices of three popular cryptocurrencies: Bitcoin, Ethereum and Polkadot. The SVR (Support vector regression) machine learning method can provide robust and accurate predictions for nonlinear and nonstationary data. This paper compares SVR radial basis functions (RBFs) and hybrid kernels based on cryptocurrency data characteristics. SVR parameters such as regularization, gamma, and epsilon can also be tuned using grid search. Our approach is tested on real-world cryptocurrency stock prices collected from Yahoo Finance. Prediction performance is measured using regression metrics like MAPE (Mean absolute percentage error) and R2 score. In our work, a MAPE value of 0.07772 and an R2 score of 0.9999 have been obtained. The results of our experiments indicate that our approach is significantly more accurate and reliable than existing methods.
This work proposes a novel methodology to help in decision making in the cryptocurrency market. Two investment strategies have been designed for Ethereum (ETH), based on predictions of the price and trend of this cryptocurrency using real data. The two Ethereum cryptocurrency prediction systems rely solely on past values of other contextual stock indices, market indicators and online trends, and ignore any technical indicators of price evolution. Real data from cryptocurrency market has been collected and processed with different feature selection methods. Applying a regression approach with Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM) networks, prediction models for the ETH price for 1, 7 and 15 days are obtained and compared. Also, support vector machine (SVM) is applied to predict the ETH price trend by applying a classification approach. In both approaches, sentiment analysis has been included to check its effect on the prediction results. The reliability of these prediction models in the current market has been evaluated by designing two original knowledge-based investment strategies. They are tested over two different time periods with real cryptocurrency market data. The results show that it is possible to generate up to 5.16 profit factor with few operations using these models. Furthermore, adding sentiment analysis has shown to have little influence. In this way, we contribute to the advancement of our knowledge of this volatile and still young cryptocurrency market, and specifically of the evolution of Ethereum and the factors that can influence its behavior.
Ksenija Doroslovački, Nikola Gradojević, Albane Tarnaud
This article develops a method for extracting information related to the underlying stock and cryptocurrency market sentiment from European put and call option prices. We study the evolution of market sentiment and predictability of prices in the S&P 500 index and Bitcoin (BTC/USD) futures markets during the 2020–2022 period. Several innovative temporal entropic and nonentropic measures of market sentiment based on a pessimistic, a market consensus, and an optimistic view are proposed in our nonlinear forecasting models. We show that these measures have significant predictive power for future spot prices at longer forecast horizons, where they statistically and economically outperform alternative models. We also find that the BTC/USD market is more susceptible to extreme sentiments reflected in demand-based shocks, while the information regarding the degree of pessimism in relation to the market consensus is more useful in forecasting the spot S&P 500 index movements in the presence of systemic shocks.
Cryptocurrency market has striking development and aims to build open, transparent and efficient financial market. More applications are tried to build on blockchain and using cryptocurrency in multiple scenarios, and DeFi ecosystem is set up. Many niche cryptocurrencies related to DeFi market may have wide utility and demand in the future. Therefore, more research is needed to focus these cryptocurrencies and try to establish sturdy price prediction. In this study, two cryptocurrencies, Binance Coin (BNB) and Huobi Tokens (HT), which are rooted in two crypto exchange platform are used for price prediction based on three machine learning models, i.e., Random Forest (RF), Long Short-Term Memory (LSTM) and Extreme Gradient Boosting (XGBoost). Four prediction windows are selected (1, 3, 7 and 30 days span). The results for different prediction window are compared and discussed. 1 day prediction window outperforms all prediction windows with the average MSE 0.0117. Additionally, RF and XGBoost outperform LSTM with lower MSE and more stable performance, while RF and XGBoost have average MSE 0.0157 and 0.0158 separately. The research tries to predict cryptocurrencies that based on relatively niche market and discuss the performance in comprehensive ways, aiming at providing novel insights into cryptocurrency price prediction.
Bitcoin price volatility fascinates both researchers and investors, studying features that influence its movement. This paper expends on previous research and examines time series data of various exogenous and endogenous factors: Bitcoin, Ethereum, S&P 500, and VIX closing prices; exchange rates of the Euro and GPB to USD; and the number of Bitcoin-related tweets per day. A period of three years (from September 2019 to September 2022) is covered by the research dataset. A two-layer framework is introduced tasked with accurately forecasting Bitcoin price. In the first layer, to account for complexities in the analyzed data, variational mode decomposition (VMD) extracts trends from the time series. In the second layer, Long short-term memory and hybrid Bidirectional long short-term memory networks were used to forecast prices several steps ahead. This work also introduced an enhanced variant of the sine cosine algorithm to tune the control parameters of VMD and both neural networks for attaining the best possible performance. The main focus is on combining VMD with modified metaheuristics to improve cryptocurrency closing value forecast. Two sets of experiments were conducted, with and without VMD. The results have been contrasted with models tuned by seven other cutting-edge optimizers. Extensive experimental outcomes indicate that Bitcoin price can be forecasted with great accuracy using selected features and time series decomposition. Additionally, the best model was analyzed, and Shapley values indicated that features such as EUR/USD exchange rates, Ethereum closing prices, and GBP/USD exchange rates, have a significant impact on forecasts.
Konstantinos I. Roumeliotis, Nikolaos D. Tselikas, Dimitrios Κ. Nasiopoulos
Cryptocurrencies are becoming increasingly prominent in financial investments, with more investors diversifying their portfolios and individuals drawn to their ease of use and decentralized financial opportunities. However, this accessibility also brings significant risks and rewards, often influenced by news and the sentiments of crypto investors, known as crypto signals. This paper explores the capabilities of large language models (LLMs) and natural language processing (NLP) models in analyzing sentiment from cryptocurrency-related news articles. We fine-tune state-of-the-art models such as GPT-4, BERT, and FinBERT for this specific task, evaluating their performance and comparing their effectiveness in sentiment classification. By leveraging these advanced techniques, we aim to enhance the understanding of sentiment dynamics in the cryptocurrency market, providing insights that can inform investment decisions and risk management strategies. The outcomes of this comparative study contribute to the broader discourse on applying advanced NLP models to cryptocurrency sentiment analysis, with implications for both academic research and practical applications in financial markets.
This study investigates the application of the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm for detecting anomalies in Bitcoin trading data. With the growing significance of Bitcoin in the financial market, identifying irregular trading patterns is crucial for maintaining market integrity and preventing market manipulation. Utilizing a dataset from Kaggle, which includes features such as date, timestamp, open, high, low, close, volume, and number of trades, the data was aggregated from minute-by-minute to hourly intervals for more manageable analysis. The DBSCAN algorithm effectively identified a primary cluster comprising 29,612 data points and flagged 2 points as anomalies, achieving a precision of 1.0, recall of 0.0068, F1-score of 0.0135, and an AUC-ROC of 0.5034. The optimal parameters, determined through sensitivity analysis, were epsilon (ε) = 0.1 and min_samples = 3, yielding the highest silhouette score of 0.21499. These results underscore the algorithm's ability to accurately label anomalies while highlighting the challenge of comprehensive anomaly detection. The study contributes to the field of financial anomaly detection by demonstrating the effectiveness of DBSCAN in analyzing high-dimensional, noisy datasets. It also addresses gaps in the literature regarding the application of density-based clustering methods to Bitcoin trading data. Despite its contributions, the study acknowledges limitations, such as potential data aggregation impact and the need for further validation with different datasets. Future research directions include integrating additional features like social media sentiment and exploring hybrid approaches that combine supervised and unsupervised methods.
This study investigates the relationship between trading volume and Bitcoin price movements using Pearson and Spearman correlation methods. The aim is to determine if trading volume can reliably predict Bitcoin price changes. Using a comprehensive dataset of daily Bitcoin prices and trading volumes, various statistical techniques were employed. Pearson and Spearman correlation analyses revealed very weak and statistically insignificant relationships, with correlation coefficients of -0.023788 and 0.021093, respectively. Linear regression analysis further supported these findings, showing an insignificant regression coefficient for trading volume and a very low R-squared value of 0.000566. Volatility analysis, measured by the standard deviation of daily returns, demonstrated high price volatility, consistent with the cryptocurrency market's nature. This volatility is influenced by factors such as market sentiment, regulatory developments, and macroeconomic events. The study also utilized 30-day moving averages to smooth short-term fluctuations and highlight long-term trends in trading volume and closing prices, revealing underlying trends not visible in daily data. A 1-day lagged correlation analysis indicated a very weak relationship (0.008145) between trading volume on one day and price changes on the next, suggesting other factors drive price movements. Visualizations, including time series graphs, histograms, moving averages, and volatility graphs, further illustrated the lack of a clear pattern between trading volume and price changes. In conclusion, trading volume is not a significant predictor of Bitcoin price movements, highlighting the need for comprehensive analytical approaches considering multiple variables to understand and predict Bitcoin price dynamics better.
This paper evaluates the performance of the Long Short-Term Memory (LSTM) deep learning algorithm in forecasting Bitcoin and Ethereum prices during the COVID-19 epidemic, using their high-frequency price information, ranging from December 31, 2019, to December 31, 2020. Deep learning (DL) techniques, which can withstand stylized facts, such as non-linearity and long-term memory in high-frequency data, were utilized in this paper. The LSTM algorithm was employed due to its ability to perform well with time series data by reducing fading gradients and reliance over time. The obtained empirical results demonstrate that the LSTM technique can predict both Ethereum and Bitcoin prices. However, the performance of this algorithm decreases as the number of hidden units and epochs grows, with 100 hidden units and 200 epochs delivering maximum forecast accuracy. Furthermore, the performance study demonstrates that the LSTM approach gives more accurate forecasts for Ethereum than for Bitcoin prices, indicating that Ethereum is more prominent than Bitcoin. Moreover, the increased accuracy of forecasting the Ethereum price made it more reliable than Bitcoin during the COVID-19 coronavirus crisis. As a result, cryptocurrency traders might focus on trading Ethereum to increase their earnings during a crisis.
This study investigates the efficacy of ARIMA and LSTM models in predicting Bitcoin prices, emphasizing the importance of accurate price prediction for trading, risk management, and investment strategies in the volatile cryptocurrency market. The objectives are to analyze Bitcoin prices to identify underlying patterns and trends, compare the predictive performance of ARIMA and LSTM models, and provide insights into their practical applications for Bitcoin price prediction. A comprehensive dataset of Bitcoin prices from January 1, 2011, to December 31, 2023, sourced from CoinMarketCap, was used. Data preprocessing included handling missing values, removing duplicates, achieving stationarity through differencing, and normalizing data using MinMaxScaler. The ARIMA model's best-fitting parameters were identified using ACF and PACF plots, and it was trained with the statsmodels library. The LSTM model involved data preparation through windowing and train-test splitting, constructing a neural network with LSTM layers, and training using TensorFlow/Keras. Evaluation metrics included Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE), with comparisons based on accuracy and computational efficiency. The ARIMA model demonstrated impressive performance with an MAE of 2.308392356829177e-215 and an RMSE of 0.0, indicating a near-perfect fit to the training data. The LSTM model achieved an MAE of 0.00021804577826689423 and an RMSE of 0.00021916977109865863, showing robust performance in handling nonlinear and long-term dependencies. The ARIMA model excelled in computational efficiency with a training time of 2.548070192337036 seconds and a prediction time of 0.0009970664978027344 seconds, while the LSTM model required 378.69622468948364 seconds for training and 0.6859967708587646 seconds for prediction. The results highlight ARIMA's effectiveness in capturing linear trends and its suitability for short-term trading strategies, while LSTM is better for long-term investment strategies due to its ability to model complex patterns. Despite potential overfitting in ARIMA and high computational demands for LSTM, the study suggests exploring hybrid models, incorporating additional data sources, and developing advanced techniques to enhance predictive accuracy in future research.
Over the past few years the interest in trading with a decentralized or virtual form of currency has significantly increased. This has led to the rise of the cryptocurrency market in the early 2000’s. Bitcoin, the pioneer in the field, has managed to dominate this volatile and cutthroat market till this day. Cryptocurrency is based on multiple technological frameworks and success stories of high returns in a short time frame have garnered the interest of young investors as well. Similar to the traditional stock market, multiple machine learning, Artificial Intelligence, Time Series Analysis models have come up to help investors, understand trends, patterns, and derive a deeper understanding of the asset as well as the market they are investing in. Our research aims to deal with this problem using time series methods such as Auto-Regressive (AR), Moving Average (MA) and Auto-Regressive Integrated Moving Average (ARIMA) models. In our analysis, we have implemented the models using both Excel method and Python. Our metric of evaluation is Mean Absolute Percentage Error (MAPE). In our work, data was taken from a website called ‘Yahoo finance’ [1] for Bitcoin cryptocurrency for a five-year time period i.e. from 1st June, 2017 to 31st May, 2024. Our Python methodology has resulted in a MAPE of 0.16 for AR model, 0.14 for MA model and 0.11 for ARIMA model. The same has been verified using Excel and the score has been validated.