M. Eren Akbiyik, Mert Erkul, Killian Kaempf, Vaiva Vasiliauskaitė · 5 authors
Understanding the variations in trading price (volatility), and its response to exogenous information, is a well-researched topic in finance. In this study, we focus on finding stable and accurate volatility predictors for a relatively new asset class of cryptocurrencies, in particular Bitcoin, using deep learning representations of public social media data obtained from Twitter. For our experiments, we extracted semantic information and user statistics from over 30 million Bitcoin-related tweets, in conjunction with 15-minute frequency price data over a horizon of 144 days. Using this data, we built several deep learning architectures that utilized different combinations of the gathered information. For each model, we conducted ablation studies to assess the influence of different components and feature sets over the prediction accuracy. We found statistical evidences for the hypotheses that: (i) temporal convolutional networks perform significantly better than both classical autoregressive models and other deep learning-based architectures in the literature, and (ii) tweet author meta-information, even detached from the tweet itself, is a better predictor of volatility than the semantic content and tweet volume statistics. We demonstrate how different information sets gathered from social media can be utilized in different architectures and how they affect the prediction results. As an additional contribution, we make our dataset public for future research.
Bitcoin's dominance in the cryptocurrency market has only increased in recent years. However, it experiences rapid spikes and declines that creates difficulty in predicting its future behavior. Much research has been done to find efficient models that predict with high accuracy, but with limited results. The goal of this study was to determine if higher accuracy can be achieved by focusing on a broader perspective of numeric ranges as opposed to specific time series price predictions. The predictions were concentrated on reporting the expected market direction for the following hour. In using one hour interval trading data and creating discrete classes of levels of hourly changes, five different Machine Learning models were trained and tested. Except for one model, cross validation accuracy ranging from 96-100% was achieved.
Abstract This study examines intraday time series momentum in Bitcoin. Unlike stock markets, Bitcoin trades 24 h a day and therefore has not got a clear opening and closing period. Therefore, we use trading volume as a proxy for the market trading time and show that the first half‐hour positively predicts the last half‐hour return. We find that the first trading sessions with the highest volume or volatility are associated with the greatest predictability for intraday time series momentum. We also show that intraday momentum‐based trading yields substantial economic gains in terms of market timing and asset allocation, especially in periods of a market downturn in Bitcoin. Consistent with the finding in foreign exchange markets, our results also show that the Bitcoin intraday momentum is driven by liquidity provision rather than late‐informed trading.
Due to the nonlinearity and highly volatile dynamics of the price data of cryptocurrency, classic parametric models show limited success in tracking and prediction. With the rise of deep learning recently, various researches on forecasting the price of cryptocurrency using deep neural network have reported encouraging results in the cases of low volatility. In this study, we propose a hybrid approach which combines the advantages of non-stationary parametric models such as Generalized Autoregressive Conditional Heteroskedasticity (GARCH) with the nonlinear modelling potential of Long-Short Term Memory (LSTM) neural networks. The results show that our hybrid model has a similar predictive performance in terms of MSE, MAE and RMSE but higher metric scores in precision, accuracy and F1 score under optimal hyperparameters. This study reveals that the combination of parametric models like GARCH with deep neural network may come up with better results in cryptocurrency price forecasting especially in the case of highly volatile data or when short data sequences are available. Moreover, the proposed framework can be used also in other applications where high volatility and scarcity of data are the main characteristics.
Simone Casale-Brunet, Paolo Ribeca, Patrick Charles Doyle, Marco Mattavelli
Non-fungible tokens (NFTs) as a decentralized proof of ownership represent one of the main reasons why Ethereum is a disruptive technology. This paper presents the first systematic study of the interactions occurring in a number of NFT ecosystems. We illustrate how to retrieve transaction data available on the blockchain and structure it as a graph-based model. Thanks to this methodology, we are able to study for the first time the topological structure of NFT networks and show that their properties (degree distribution and others) are similar to those of interaction graphs in social networks. Time-dependent analysis metrics, useful to characterize market influencers and interactions between different wallets, are also introduced. Based on those, we identify across a number of NFT networks the widespread presence of both investors accumulating NFTs and individuals who make large profits.
Throughout numerous cryptocurrency exchanges, markets exhibit recurrent opportunities for arbitrage. There are various types of arbitrage strategies that can be carried out. Taking consideration of the pricing data from three famous exchanges (Binance, Kucoin and Coinbase) for bitcoin, the data analysis on spot prices for each timeframe helps in determining the buy market and sell market. Apart from the analysis, the profits generated by performing arbitrage using one of the discussed strategies are quantified. The implemented algorithm can be modified to suit the trading strategy. Finally, the future scope, as well as the issues and potential risks involved in these strategies, are discussed.
Barbara Będowska-Sójka, Agata Kliber, Aleksandra Rutkowska
We try to establish the commonalities and leadership in the cryptocurrency markets by examining the mutual information and lead-lag relationships between Bitcoin and other cryptocurrencies from January 2019 to June 2021. We examine the transfer entropy between volatility and liquidity of seven highly capitalized cryptocurrencies in order to determine the potential direction of information flow. We find that cryptocurrencies are strongly interrelated in returns and volatility but less in liquidity. We show that smaller and younger cryptocurrencies (such as Ripple's XRP or Litecoin) have started to affect the returns of Bitcoin since the beginning of the pandemic. Regarding liquidity, the results of the dynamic time warping algorithm also suggest that the position of Monero has increased. Those outcomes suggest the gradual increase in the role of privacy-oriented cryptocurrencies.
Cryptocurrency is in great demand today and there is pretty much investment in cryptocurrencies by the investors. There are more than 6000 cryptocurrencies all over the world, which clearly shows that cryptocurrency is a growing investment market. For this reason, investors having ordinary income invest in promising cryptocurrencies with a low market value. However, these investors are often unconsciously investing and making losses. At this point, sensible investments can be made using data analysis methods based on deep learning. Therefore, this study aims to analyze promising cryptocurrencies with deep learning methods. Five promising cryptocurrencies were analyzed with the ensembles of LSTM and single-based LSTM networks. This study revealed that ensembles of LSTM network do not always provide better accuracy performance than the single-based LSTM network in the analysis of promising cryptocurrencies. In other words, these two deep learning methods can be employed to obtain reliable analysis results in promising cryptocurrencies.
Kevin E. Martin, Mohamed Rahouti, Moussa Ayyash, Izzat Alsmadi
Abstract The vast majority of digital currency transactions rely on a blockchain framework to ensure quick and accurate execution. As such, understanding how a blockchain works is vital to understanding the dynamics of cryptocurrency operations. One of the key benefits of this type of system is the exhaustive records captured in a given marketplace. The interwoven movement between agents can effectively be expressed as a graph via the extraction of historical data from the blockchain. By looking at a specific blockchain as an interaction of its agents, network representation learning can be leveraged to examine these relationships. Furthermore, the analysis of a graph structure can be enhanced through the application of modern and sophisticated machine learning techniques. Leveraging the automated nature of these methods can create meaningful observations of the input network. In this paper, we utilize several machine learning models to detect anomalous transactions in various digital currency markets. We find that supervised learning techniques yield encouraging results, whereas unsupervised learning techniques struggle more with the classification.
Risk management and prediction of market losses of cryptocurrencies are of notable value to risk managers, portfolio managers, financial market researchers and academics. One of the most common measures of an asset’s risk is Value-at-Risk (VaR). This paper evaluates and compares the performance of generalized autoregressive score (GAS) combined with heavy-tailed distributions, in estimating the VaR of two well-known cryptocurrencies’ returns, namely Bitcoin returns and Ethereum returns. In this paper, we proposed a VaR model for Bitcoin and Ethereum returns, namely the GAS model combined with the generalized lambda distribution (GLD), referred to as the GAS-GLD model. The relative performance of the GAS-GLD models was compared to the models proposed by Troster et al. (2018), in other words, GAS models combined with asymmetric Laplace distribution (ALD), the asymmetric Student’s t-distribution (AST) and the skew Student’s t-distribution (SSTD). The Kupiec likelihood ratio test was used to assess the adequacy of the proposed models. The principal findings suggest that the GAS models with heavy-tailed innovation distributions are, in fact, appropriate for modelling cryptocurrency returns, with the GAS-GLD being the most adequate for the Bitcoin returns at various VaR levels, and both GAS-SSTD, GAS-ALD and GAS-GLD models being the most appropriate for the Ethereum returns at the VaR levels used in this study.
Yuzhi Cai, Thanaset Chevapatrakul, Danilo V. Mascia
Abstract We shed light on how the price explosivity characterising Bitcoin and other major cryptocurrencies is triggered, by employing the Quantile Self-Exciting Threshold Autoregressive (QSETAR) model. Our results for Bitcoin, Ripple, and Stellar reveal that the explosive behaviour originates from the extreme upper tails of the return distributions following a price increase in the preceding day. We do not find evidence of explositivity in the price of Litecoin.
In this study, we predicted the log returns of the top 10 cryptocurrencies based on market cap, using univariate and multivariate machine learning methods such as recurrent neural networks, deep learning neural networks, Holt’s exponential smoothing, autoregressive integrated moving average, ForecastX, and long short-term memory networks. The multivariate long short-term memory networks performed better than the univariate machine learning methods in terms of the prediction error measures.
In this paper, we conduct a fast calibration in the jump-diffusion model to capture the Bitcoin price dynamics, as well as the behavior of some components affecting the price itself, such as the risk of pitfalls and its ambiguous effect on the evolution of Bitcoin’s price. In addition, in our study of the Bitcoin option pricing, we find that the inclusion of jumps in returns and volatilities are significant in the historical time series of Bitcoin prices. The benefits of incorporating these jumps flow over into option pricing, as well as adequately capture the volatility smile in option prices. To the best of our knowledge, this is the first work to analyze the phenomenon of price jump risk and to interpret Bitcoin option valuation as “exceptionally ambiguous”. Crucially, using hedging options for the Bitcoin market, we also prove some important properties: Bitcoin options follow a convex, but not strictly convex function. This property provides adequate risk assessment for convex risk measure.
Motivated by the unique transaction cost structure of the cryptocurrency (CC) market1, this study investigates the phenomenon of liquidity commonality across a sample of 53 CCs. The study employs the google search volume index (GSVI) measure to capture the retail investor’s attention towards the CC market. Using the quantile regression method, we document the liquidity dynamics of CCs that is contrasting to other asset classes, and is ascribed to its unique transaction cost structure. In view of the relatively high liquidity commonality levels found in the CC market, this paper sounds a note of caution to retail investors on episodic non-availability of liquidity in CCs.
The structural variations related to legal tender of cryptocurrencies operating across the world markets has been instable since their inception. There have been many changes incorporated for their status to be recognized as a legal financial instrument for investment purposes over the virtual financial markets. The concept of anomalies associated with the popular efficient market hypotheses given by Eugene Fama existed for Bitcoin and few other cryptocurrencies over a period of time. The present work attempts to contribute to the existing literature on cryptocurrency studies. A focused investigation has been carried for cryptocurrencies to find different behaviour of returns on these currencies over a varied response from various countries with respect to their permissible manoeuvres. The study has used independent sample t-test, one-way ANOVA and dummy regression analysis to examine the day of the week effect for a time period between 2014–2020 split into multiple sub-periods. Anomalies have been found for cryptocurrencies across multiple sub-periods with varied magnitude.
More and more people are entering the cryptocurrency market after Bitcoin (BTC) soared to nearly USD 20,000 in 2017. To promote the development of information technology and cryptocurrency marketing, various computerized systems integrating information technology with investment and financing are innovated continuously. In this study, the daily cryptocurrency prices were input to Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM); and the developing trend plots were drawn to predict and analyze the future cryptocurrency prices through deep learning. Finally, the business practices of cryptocurrency investment were modularized based on High-Level Fuzzy Petri Nets (HLFPNs) to make a better investment decision so that all investors can use this decision system to quickly understand the future cryptocurrency trend. The experimental results have shown that this decision system can provide effective investment information to achieve investors’ personal financial goals with the expectation of improving financial situations.
The clustering of cryptocurrencies - as an emerging field in investment management - is the main topic of this research. Applying the information-based distance matrices, we clustered the 30 most valuable cryptocurrencies. Then, we identified the most influential clustering by the concept of Minimum Spanning Tree (MST) and the centrality measures of graph theory. A second-order clustering, which is defined as the clustering of hierarchical clusterings, is applied to cluster 56 dendrograms. Using the most influential clustering, we identified the main clusters of cryptocurrencies and sub-clusters. The results show that the clustering composition of cryptocurrencies changed at the period I (before COVID-19) and II (pandemic time).
Sep 30, 2021·Vestnik Voronezhskogo gosudarstvennogo universiteta Ser Ekonomika i upravlenie = Proceedings of Voronezh State University Series Economics and Management
Dmitry А. Endovitsky, Вячеслав Владимирович Коротких
Introduction. Digital financial assets are a relatively new phenomenon. More and more, they include virtual currencies, and in particular cryptocurrencies. Both regulators and financial market players are becoming increasingly interested in such assets. Cryptocurrencies have no intrinsic value, and this encourages scientific studies on the problem of price formation and risk management associated with cryptocurrency operations. Most papers on the problem lack a systematic ap-proach and do not provide solutions to a large number of fundamental issues. Purpose. The purpose of our study was to develop a method for the risk analysis of operations with digital financial assets, namely cryptocurrencies. Methodology. In our study, we used parametric methods of data analysis and ma-chine learning methods, description, analysis, synthesis, induction, deduction, comparison, and grouping method. The sample was accumulated between April 2013 and April 2021 and included cryptocurrencies with the market capitalization of over 1 million USD. Results. The study determined the common risk factors for the cryptocurrency market. The risk factors are presented as linear combinations of returns of subsets of cryptocurrencies with dynamically changing weight coefficients. The risk factors were formed based on the market information, which included the price of the cryptocurrency, the trading volume, and its market capitalization. Conclusions. The study demonstrated that the cryptocurrency market is suscepti-ble to market anomalies common to traditional financial asset markets. In addition to the risk factors based on the market capitalization of cryptocurrencies (the size) and their aggregate profitability (the momentum), the article presents statistically relevant risk factors which reflect the growth rate of the market capitalization and the level of illiquidity of cryptocurrencies. In order to explain the market anomalies and the arbitrary strategies based on them, the article presents several factor models of cryptocurrency price formation. These models can be used to develop an in-tegrated approach to the risks associated with operations with digital financial assets.
The paper is devoted to studying the bitcoin blockchain as a new global phenomenon in monetary economics, which requires comprehending from the economic theory view - a self-regulating system of decentralized emission without participation of a central monetary authority. Mathematical modelling is the instrument of this studying. The author has analyzed the Bitcoin system parameters that determine dynamics of a self-regulating emission mechanism. This mechanism operates in a peer-to-peer computer network and provides a smooth increasing of the "money supply" with a gradually decreasing rate of growth. The limit of this growth is determined by maximal volume 21 million BTC. Self-regulation is implemented through negative feedback between changes of control parameters (the target interval for the hash function values and the Bitcoin Difficulty level) and the speed of mining process. Control parameters depend on the real speed deviations from the target value. This mechanism provides a stable mining speed and determines annual rate of emission. The author suggests a spline-function for describing the annual rate of the cryptocurrency emission in accordance with the Proof-of-Work protocol in the Bitcoin blockchain algorithm. This spline-function gives possibility to find a monetary rule for annual rate of emission. The author in the paper proposes to call this monetary rule by the name of the Bitcoin system inventor - Nakamoto Monetary Rule. The Nakamoto Monetary Rule could be seen as the first example of a programmable monetary rule of the decentralized emission algorithm on the basis of blockchain technology. Central banks could use a similar approach, with the necessary modifications, to develop their programmable monetary rules for Central Bank Digital Currencies (CBDCs) emission based on DLT or blockchain technology
The blockchain technology introduced by bitcoin, with its decentralised peer-to-peer network and cryptographic protocols, provides a public and accessible database of bitcoin transactions that have attracted interest from both economics and network science as an example of a complex evolving monetary network. Despite the known cryptographic guarantees present in the blockchain, there exists significant evidence of inconsistencies and suspicious behavior in the chain. In this paper, we examine the prevalence and evolution of two types of anomalies occurring in coinbase transactions in blockchain mining, which we reported on in earlier research. We further develop our techniques for investigating the impact of these anomalies on the blockchain transaction network, by building networks induced by anomalous coinbase transactions at regular intervals and calculating a range of network measures, including degree correlation and assortativity, as well as inequality in terms of wealth and anomaly ratio using the Gini coefficient. We obtain time series of network measures calculated over the full transaction network and three sub-networks. Inspecting trends in these time series allows us to identify a period in time with particularly strange transaction behavior. We then perform a frequency analysis of this time period to reveal several blocks of highly anomalous transactions. Our technique represents a novel way of using network science to detect and investigate cryptographic anomalies.
Exploring the evolution of the transaction network is very important for analyzing anonymous transaction behavior of encrypted currency and avoiding illegal crimes. However, with the rapid growth of cryptocurrency transactions in recent years, the traditional static analysis models tend to ignore parameter changes in the evolution process. To solve this problem, constructing time-varying network model is an effective scheme. The main challenge of turning the massive statically stored transaction data into dynamical sequential network is to design a set of suitable network model. This paper took Bitcoin system as example, combined complex network evolution theory, proposed a weighted variable directed bipartite graph (VW-DBG) model. Initially, we defined the transaction weights and the influence inditcator of nodes, and accordingly introduced the node deletion mechanism to Bitcoin transaction network analysis for the first time. Moreover, information entropy indicators was defined to screen key periods in the evolution. In addition, we analyzed the dynamic and static indicators of real Bitcoin transaction network under different observation time using this model, and revealed the pow-law distribution in evolutionary networks.
Abstract In this article, we describe what cryptocurrency is, how it works, and how it relates to familiar conceptions of and questions about money. We then show how normative questions about monetary policy find new expression in Bitcoin and other cryptocurrencies. These questions can play a role in addressing not just what money is, but what it should be. A guiding theme in our discussion is that progress here requires a mixed approach that integrates philosophical tools with the purely technical results of disciplines like computer science and economics.