Cryptocurrencies are significant improvements in the digital age that have changed the way we think about money. The first cryptocurrency was Bitcoin, introduced in 2009 and was created by Nakamoto. Due to their potential ups and downs, many people now think that cryptocurrencies are appropriate for use as an investment instrument, especially millennials who are attracted to higher-risk investment alternatives. A number of different investing options such as cryptocurrencies, gold, and other conventional assets like equities and bonds have unique characteristics and advantages. It’s essential for investors to understand the similarities and differences between cryptocurrencies and other assets in order to create diversified portfolios. In this study, the optimum portfolio will be constructed using Bitcoin, Gold, LQ45 Index, and ABF IBI as the representative of Indonesia Bond Index. Mean-Variance Optimization will be used as an asset allocation method, and will be compared to the other methods such as Risk Parity, 60/40 Portfolio, and Equally Weighted to find a better risk-adjusted return. The Sharpe ratio analysis is used to evaluate the portfolio performance resulting from every method. The investment strategy will be simulated to know which strategy will result the best total return in the end of simulation period. According to risk, return, and the Sharpe ratio, Bitcoin could perform better than gold, LQ45, and ABF IBI. Furthermore, the Mean-Variance Optimization resulted the highest Sharpe ratio compared to the other methods. The optimal weight from the portfolio construction using Mean-Variance Optimization allocated 53% to ABFI index, 40% to Bitcoin, and 7% to gold, which resulted 48.2% portfolio return, 40.44% portfolio risk, and 1077.8% Sharpe ratio. From the investment strategy simulation, the quarterly rebalancing strategy was found to be the best strategy with the total return 223.36%.
Due to many factors, Bitcoin has experienced huge price fluctuations since its emergence, and it has received extensive attention. Forecasting the price of bitcoin is of great significance for investors and for the country's future development. This paper collects the data of bitcoin price and indicator that may affect the price, and then use random forest algorithm for feature selection to remove all nonessential indicators. Then, CNN-Bi-LSTM-Attention model is built to train the data and predict the price of bitcoin. Finally, this model is compared with other models. It can be found that this model has higher prediction accuracy and better prediction effect than traditional models such as LSTM and CNN-LSTM.
Dr.Alamelu Mangai Jothidurai, Pratheek D Kanchan, Rahul Raj
Abstract: Cryptocurrency price prediction is a challenging task due to the high volatility and uncertainty of the market. Machine learning techniques can provide useful insights and forecasts for investors and traders. In this paper, we propose a novel approach for cryptocurrency price prediction using machine learning models and sentiment analysis. We collect historical price data of Bitcoin from yahoo business. We then apply various machine learning models, such as LSTM for the price prediction of the cryptocurrency using the past data. LSTM is a type of recurrent neural network that can manage long-term dependencies and sequential data. LSTM has three gates: forget gate, input gate, and output gate, which control the flow of information in and out of the memory cell. LSTM can be implemented in Python using the Keras and TensorFlow library. In this paper, we use LSTM as one of the machine learning models for cryptocurrency price prediction. We then use the average of the next 5 days of the predicted data to implement a buy-sell call strategy that aims to maximize the profit and minimize the risk. We evaluate our framework on a popular cryptocurrency Bitcoin.
Eddie Ngai, Salwani Abdullah, Salwani Abdullah, Mohd Zakree Ahmad Nazri · 6 authors
Over the past few years, Bitcoin has attracted the attention of numerous parties, ranging from academic researchers to institutional investors. Bitcoin is the first and most widely used cryptocurrency to date. Due to the significant volatility of the Bitcoin price and the fact that its trading method does not require a third party, it has gained great popularity since its inception in 2009 among a wide range of individuals. Given the previous difficulties in predicting the price of cryptocurrencies, this project will be developing and implementing a time series approach-based solution prediction model using machine learning algorithms which include Support Vector Machine Regression (SVR), K-Nearest Neighbor Regression (KNN), Extreme Gradient Boosting (XGBoost), and Long Short-Term Memory (LSTM) to determine the trend of bitcoin price movement, and assessing the effectiveness of the machine learning models. The data that will be used is the close prices of Bitcoin from the year 2018 up to the year 2023. The performance of the machine learning models is evaluated by comparing the results of R-squared, mean absolute error (MAE), mean squared error (RMSE), and also through a visualization graph of the original close price and predicted close price of Bitcoin in a dashboard. Among the models compared, LSTM emerged as the most accurate, followed by SVR, while XGBoost and KNN exhibited comparatively lower performance.
This research addresses the impact of individual investors on the cryptocurrency market, focusing specifically on the development of herd behavior. Although the phenomenon of herd behavior has been studied extensively in the stock market, it has received limited research in the context of cryptocurrencies. This study aims to fill this research gap by examining the impact of liquidity and sentiment on herd behavior using the CSAD model, considering small, medium, and large cryptocurrencies. The results show different outcomes for cryptocurrencies of different sizes, consistently demonstrating that the herding effect is more pronounced under conditions of lower liquidity, as determined by the turnover volume and liquidity ratio of cryptocurrencies. Proxy measures such as the Twitter Hedonometer and CBOE VIX were used to measure investor sentiment and show the prevalence of herding behavior in optimistic times for all cryptocurrencies, regardless of their market capitalization. Consequently, this study provides valuable insights into the manifestation of herd behavior in the cryptocurrency market and highlights the importance of liquidity and sentiment as influencing factors. These findings improve our understanding of investor behavior and provide guidance to market participants and policymakers on how to effectively manage the risks associated with herd effects.
Markus Frohmann, Manuel Karner, Said Khudoyan, Robert F. Wagner · 5 authors
Recently, various methods to predict the future price of financial assets have emerged. One promising approach is to combine the historic price with sentiment scores derived via sentiment analysis techniques. In this article, we focus on predicting the future price of Bitcoin, which is currently the most popular cryptocurrency. More precisely, we propose a hybrid approach, combining time series forecasting and sentiment prediction from microblogs, to predict the intraday price of Bitcoin. Moreover, in addition to standard sentiment analysis methods, we are the first to employ a fine-tuned BERT model for this task. We also introduce a novel weighting scheme in which the weight of the sentiment of each tweet depends on the number of its creator’s followers. For evaluation, we consider periods with strongly varying ranges of Bitcoin prices. This enables us to assess the models w.r.t. robustness and generalization to varied market conditions. Our experiments demonstrate that BERT-based sentiment analysis and the proposed weighting scheme improve upon previous methods. Specifically, our hybrid models that use linear regression as the underlying forecasting algorithm perform best in terms of the mean absolute error (MAE of 2.67) and root mean squared error (RMSE of 3.28). However, more complicated models, particularly long short-term memory networks and temporal convolutional networks, tend to have generalization and overfitting issues, resulting in considerably higher MAE and RMSE scores.
The most popular cryptocurrency used worldwide is bitcoin. Many everyday folks and investors are now investing in bitcoin. However, it becomes quite difficult to evaluate or foresee the price of bitcoin. The price of bitcoin is extremely difficult to forecast due to its swings. By this point, machine learning has developed a number of models to examine the price behaviour of bitcoin using time series data. The digital money, a different type of payment developed utilising encryption methods, is difficult to forecast. By utilising encryption technology, cryptocurrencies may act as both a medium of exchange and a virtual accounting system. To estimate the values of a future time sequence, this work introduces a deep learning-based technique for time series forecasting that treats the current data as time series and extracts the key traits of the past. To overcome the shortcomings of conventional production forecasting, three algorithms-auto-regressive integrated moving averages (ARIMA), long-short-term memory (LSTM) network, and FB-prophet-were investigated and contrasted. We compared the models using historical bitcoin data of past eight years, from 2012 to 2020. The “FB-prophet” model, which is significant, catches variation that might draw attention and avert possible problems.
We propose an ensemble method to improve the generalization performance of trading strategies trained by deep reinforcement learning algorithms in a highly stochastic environment of intraday cryptocurrency portfolio trading. We adopt a model selection method that evaluates on multiple validation periods, and propose a novel mixture distribution policy to effectively ensemble the selected models. We provide a distributional view of the out-of-sample performance on granular test periods to demonstrate the robustness of the strategies in evolving market conditions, and retrain the models periodically to address non-stationarity of financial data. Our proposed ensemble method improves the out-of-sample performance compared with the benchmarks of a deep reinforcement learning strategy and a passive investment strategy.
Mingzhe Wei, Ioannis Kyriakou, Georgios Sermpinis, Charalampos Stasinakis
Abstract This study explores the effectiveness of technical and fundamental analysis in predicting and trading the returns of 12 cryptocurrencies, namely Bitcoin, Ethereum, Ripple, Dash, Cardano, Avalanche, Binance Coin, Dogecoin, Polkadot, Litecoin, Terra and Solana. A universe of 7846 technical rules, five log moving average‐based ratios and 59 fundamental factors are used to test predictability and profitability through the Lucky Factors methodology and Superior Predictive Ability test. We observe predictability for a small set of technical and fundamental rules, while only the short‐term log moving average‐based ratio and Hashrate Index demonstrate genuine in‐sample and out‐of‐sample profitability. Our findings question the value of both technical and fundamental analysis on cryptocurrencies.
I Made Gede Abandi Semeru, Yunieta Anny Nainggolan
Abstract : In forming their portfolios, investors should analyze the risk and return of each investment instrument. This is aimed at preventing investors from speculating and gambling with their investments. Conducting an investment portfolio optimization study on LQ-45 stock index, government bond, USD, gold, and Bitcoin can provide valuable insights due to unique market characteristics in Indonesia. This research analyzes the formation of investment instruments over the last 60 months, specifically from January 2018 to December 2022. The research method used in this study is quantitative research aimed at selecting several investment instruments for a portfolio in Indonesia. The portfolio aims to minimize risk and maximize return using the Markowitz method, also known as the optimal portfolio. To fulfill the objectives of this research, data on the prices of each instrument are required. An optimal portfolio can be obtained by combining two instruments: 18% bitcoin and 82% gold. This optimal portfolio can achieve an expected return of 1.29% with a risk level of 5.15%. Considering a risk-free rate of 0.375%, this portfolio forms a slope of 0.1775, which is the largest slope formed between the combination of risk-free instruments and risky portfolios. Investors should allocate their funds more wisely, considering not only the highest return but also the associated risk. High returns often come with high risks, so investors need to assess the risk-return trade-off before making investment decisions.
Cryptocurrency has become a popular asset in global financial markets, meaning that individual investors and asset management companies worldwide are considering this new investment class. The main contribution of this research is to address an intra-day forecasting problem with hourly granularity by comparing deep network architectures, including ones with attention mechanisms for the Ethereum intrinsic cryptocurrency (ETH). Since variations on the deep learning model parameter values may also introduce variability in the results produced by the models, different statistical validations were considered part of the comparison process. Finally, this work shows that the Temporal Convolutional Network model (TCN) outperformed other architectures considered for a short-term forecast period in terms of processing time. The TCN deep learning model is also amongst the most accurate models, using an auto-regressive integrated moving average model (ARIMA) as a baseline.
I.sibel KERVANCI, Mehmet Fatih Akay, Eren Özceylan
Bitcoin has high price fluctuations, which involve high risks and high return rates for investors. These high earnings have attracted the attention of investors. This paper proposes a new model for Bitcoin price prediction that effectively reduces prediction error. Hyperparameter optimization methods such as Bayesian optimization (BO), random search and grid search with Long Short-Term Memory (LSTM), Gated Repetitive Unit (GRU), and hybrid LSTM-GRU utilised. Models with BO achieved better results than others. To improve each model's results with BO; Gradient Incremental Regression Trees (GBRT), Gaussian Process (GP), Random Forest (RF) and Extra Trees (ET) were applied to optimizers and corresponding surrogate functions. Evaluating the effects of hyper-parameter values on the problem for each method contributes to the parameter selection process for similar prediction problems. To increase comparability in the literature, Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE) and Mean Square Error (MSE) were used. There is a least one hyper-parameter combination, which produces a result close to the best value for each model when the results obtained from the experiments are interpreted. BO with hybrid LSTM-GRU outperformed all methods in this paper and the examined literature for the value of RMSE, MSE, and MAE.
Д. В. Фирсов, S. N. Silvestrov, Nikolay Kuznetsov, Evgeny V. Zolotarev · 5 authors
This paper identifies hidden patterns between trading volumes and the market value of an asset. Based on open market data, we try to improve the existing corpus of research using new, innovative neural network training methods. Dividing into two independent models, we conducted a comparative analysis between two methods of training Proximal Policy Optimization (PPO) models. The primary difference between the two PPO models is the data. To showcase the drastic differences the PPO model makes in market conditions, one model uses historical data from Binance trading history as a data sample and the trading pair BNB/USDT as a predicted asset. Another model, apart from purely price fluctuations, also draws data on trading volume. That way, we can clearly illustrate what the difference can be if we add additional markers for model training. Using PPO models, the authors conduct a comparative analysis of prediction accuracy, taking the sequence of BNB token values and trading volumes on 15-minute candles as variables. The main research question of this paper is to identify an increase in the accuracy of the PPO model when adding additional variables. The primary research gap that we explore is whether PPO models specifically trained on highly volatile assets can be improved by adding additional markers that are closely linked. In our study, we identified the closest marker, which is a trading volume. The study results show that including additional parameters in the form of trading volume significantly reduces the model's accuracy. The scientific contribution of this research is that it shows in practice that the PPO model does not require additional parameters to form accurately predicting models within the framework of market forecasting. Doi: 10.28991/ESJ-2023-07-04-012 Full Text: PDF
Pairs trading is a popular quantitative trading strategy with the advantage of a similarity in price movement to financial assets. Assuming that the price spreads of trading pairs are mean-reverting, this strategy exploits the disequilibrium in financial markets to find arbitrage investment opportunities. Pairs trading has been widely applied to stock, ETF, and commodity markets. However, the effectiveness of this method for cryptocurrency markets has yet to be properly explored. Therefore, we examine the profitability of pairs trading for 26 cryptocurrencies traded on the Binance exchange at high frequencies of 1, 5, and 60 min. In addition to the traditional statistical methods of distance, correlation, cointegration, and stochastic differential residual (SDR), we focus on two evolutionary algorithms: genetic algorithm (GA) and non-dominated sorting genetic algorithm II (NSGA-II). During the 79-trading-day period from 11 January to 31 March 2018, NSGA-II showed the best results at all frequencies, with an average return of 2.84%. Among the statistical models, SDR ranks first, whereas Correlation ranks last, with average returns of 1.63% and −0.48%, respectively. The z-test results show that the models are statistically significantly different. We propose NSGA-II as the best candidate for use in pairs trading strategies in cryptocurrency markets.
This study focuses on using the Neural Prophet framework to forecast Bitcoin prices accurately. By analyzing historical Bitcoin price data, the study aims to capture patterns and dependencies to provide valuable insights and predictive models for investors, traders, and analysts in the volatile cryptocurrency market. The Neural Prophet framework, based on neural network principles, incorporates features such as automatic differencing, trend, seasonality considerations, and external variables to enhance forecasting accuracy. The model was trained and evaluated using performance metrics such as RMSE, MAE, and MAPE. The results demonstrate the model's effectiveness in capturing trends and predicting Bitcoin prices while acknowledging the challenges posed by the inherent volatility of the cryptocurrency market.
The volatility and complexity of Bitcoin make it a challenging task to accurately predict its price. While past research has implemented machine learning to enhance the precision of Bitcoin price prediction, limited attention has been given to examining the viability of employing diverse modeling techniques to datasets with varying data structures and dimensional attributes. In order to forecast Bitcoin prices using machine learning techniques at different intervals, this study initiates by categorizing Bitcoin prices into daily prices and highfrequency prices. This project aims to predict the price of Bitcoin using machine learning techniques, specifically the Random Forest Classifier algorithm.
Mansour Davoudi, Mina Ghavipour, Morteza Sargolzaei-Javan, Saber Dinparast
<title>Abstract</title> This study focuses on analyzing four of the most significant cryptocurrencies in the field of decentralized storage, including Filecoin, Arweave, Storj, and Siacoin. Our method consists of three main components: Network Analysis, Textual Analysis, and Market Analysis. Network Analysis involves identifying relevant entities associated with the target cryptocurrencies to construct a network of entities. During this component, the embeddings of each entity are then extracted using node2vec which are fed into a convolutional neural network. In the second component, Textual Analysis, we first employ the T5 summarization model to encapsulate the content of related news articles. Subsequently, by utilizing the FinBert model the sentiment of news articles and tweets associated with the identified entities are extracted. We then use transformer encoders to process the resulting feature vectors. Ultimately, similar to the Textual component, by leveraging the transformer encoders the financial market information of target cryptocurrencies is evaluated during the Market Analysis component. As the final step, the outputs of these components are combined to predict the price trend of the target cryptocurrencies within a specified time frame. The proposed model’s accuracy in forecasting the future price trend of Filecoin, Storj, Arweave, and Siacoin is 76%, 83%, 61%, and 74% respectively.
In this dissertation, I investigate the existing smart contract problems that limit cognitive abilities. I use Taylor's serious expansion, polynomial equation, and fraction-based computations to overcome the limitations of calculations in smart contracts. To prove the hypothesis, I use these mathematical models to compute complex operations of naive Bayes, linear regression, decision trees, and neural network algorithms on Ethereum public test networks. The smart contracts achieve 95\% prediction accuracy compared to traditional programming language models, proving the soundness of the numerical derivations. Many non-real-time applications can use our solution for trusted and secure prediction services.
Many researchers agree that sentiment analysis can improve the performance of quantitative trading models. We develop two off-the-shelf solutions for analyzing the sentiments of cryptocurrency-related social media posts. First, we posttrain and fine-tune a Twitter-oriented model based on the bidirectional encoder representations from transformers (BERT) architecture, BERTweet, on the cryptocurrency domain, resulting in CryptoBERT. Second, we generate the language-universal cryptocurrency emoji (LUKE) sentiment lexicon and prediction pipeline, utilizing the sentiment of emojis prevalent in social media. CryptoBERT is highly accurate, while LUKE is suitable for non-English posts, thus allowing for direct classification and noisy label generation in less popular languages. Our research can help cryptocurrency investors develop trading software supported by sentiments mined from social media.
Vera Mita Nia, Ossi Ferli, Irvan Novikri, Roy Sembel · 5 authors
Increasing market capitalization is characterized by high volatility but doesn’t have the ability and potential for monetary function, Crypto world eventually shifted into the most attractive investment in the digital economy. Numerous published studies have required some improvement in the consistent relationship between commodities and financial assets and the authors proposed an alternative assessment with demonstrating the relationship between the trading volume activity of the most traded cryptocurrency in Indonesia (i.e., Ethereum) and other investment assets in Indonesia such as market indexes, rupiah exchange rate against the dollar, and gold, and related to cryptocurrencies in Indonesia which observed in over the last three years. A Var model as a quantitative and statistical approach introduced and tested the stationary data with significancy value to identify the level of acceptance model. Consistency results from previous studies where Ethereum has the largest average return but higher risk and Gold as safer investment, ultimately diversification of the investment portfolio is suggested considering the degree of risk aversion. ABSTRAK Kapitalisasi pasar yang meningkat ditandai dengan volatilitas yang tinggi namun tidak memiliki kemampuan dan potensi fungsi moneter, dunia Crypto akhirnya bergeser menjadi investasi paling menarik di ekonomi digital. Sejumlah penelitian yang diterbitkan memerlukan beberapa perbaikan dalam hubungan yang konsisten antara komoditas dan aset keuangan dan penulis mengusulkan penilaian alternatif dengan menunjukkan hubungan antara aktivitas volume perdagangan mata uang kripto yang paling banyak diperdagangkan di Indonesia (yaitu, Ethereum) dan aset investasi lainnya di Indonesia seperti indeks pasar, nilai tukar rupiah terhadap dolar, dan emas, serta terkait cryptocurrency di Indonesia yang diamati selama tiga tahun terakhir. Model Var sebagai pendekatan kuantitatif dan statistik memperkenalkan dan menguji data stasioner dengan nilai signifikansi untuk mengidentifikasi tingkat penerimaan model. Hasil konsistensi dari studi sebelumnya di mana Ethereum memiliki pengembalian rata-rata terbesar tetapi risiko lebih tinggi dan Emas sebagai investasi yang lebih aman, pada akhirnya diversifikasi portofolio investasi disarankan dengan mempertimbangkan tingkat penghindaran risiko.
Blockchain is the foundation of cryptocurrency and enables decentralized transactions through its immutable ledger. The technology uses hashing to ensure secure transactions and is becoming increasingly popular due to its wide range of applications. Python is a performant, secure, scalable language well-suited for blockchain applications. It provides developers free tools for faster code writing and simplifies crypto analysis. Python allows developers to code blockchains quickly and efficiently as it is a completely scripted language that does not require compilation. Different models such as SVR, ARIMA, and LSTM can be used to predict cryptocurrency prices, and many Python packages are available for seamlessly pulling cryptocurrency data. Python can also create one's cryptocurrency version, as seen with Facebook's proposed cryptocurrency, Libra. Finally, a versatile and speedy language is needed for blockchain applications that enable chain addition without parallel processing, so Python is a suitable choice.
Bitcoin is one of the cryptocurrencies that has gained popularity in recent years. Previous studies have shown that closing price alone is not enough to forecast its future level, and other price-related features are necessary to improve forecast accuracy. We introduce a new set of time series and demonstrate that a subset is necessary to improve directional accuracy based on a machine learning ensemble. In our experiments, we study which time series and machine learning algorithms deliver the best results. We found that the most relevant time series that contribute to improving directional accuracy are open, high, and low, with the largest contribution of low in combination with an ensemble of a gated recurrent unit network and a baseline forecast. The relevance of other Bitcoin-related features that are not price-related is negligible. The proposed method delivers similar performance to the state of the art when observing directional accuracy.
This research aims to analyze and explain the importance of diversification benefit of cryptocurrencies and its nature in accordance with its relation with other financial assets before and especially during the Covid-19 pandemic era. This paper will help investors to understand that how to manage a portfolio of cryptocurrencies in parallel with other financial assets and mainly cryptocurrencies since they were a safe investment option during the pandemic period due to the good defense these digital currencies activated against the covid-19 shock back in 2020. Paper used DCC-GARCH model to examine the safety of Bitcoin and Ethereum with financial market of S & P 500 and FTSE100.
Considering that both the entropy-based market information and the Hurst exponent are useful tools for determining whether the efficient market hypothesis holds for a given asset, we study the link between the two approaches. We thus provide a theoretical expression for the market information when log-prices follow either a fractional Brownian motion or its stationary extension using the Lamperti transform. In the latter model, we show that a Hurst exponent close to 1/2 can lead to a very high informativeness of the time series, because of the stationarity mechanism. In addition, we introduce a multiscale method to get a deeper interpretation of the entropy and of the market information, depending on the size of the information set. Applications to Bitcoin, CAC 40 index, Nikkei 225 index, and EUR/USD FX rate, using daily or intraday data, illustrate the methodological content.