Cryptocurrency price prediction using traditional statistical and machine-learning techniques: A survey

Cryptocurrencies are decentralized electronic counterparts of government-issued money. The first and best-known cryptocurrency example is bitcoin. Cryptocurrencies are used to make transactions anonymously and securely over the internet. The decentralization behavior of a cryptocurrency has radically reduced central control over them, thereby influencing international trade and relations. Wide fluctuations in cryptocurrency prices motivate the urgent requirement for an accurate model to predict its price. Cryptocurrency price prediction is one of the trending areas among researchers. Research work in this field uses traditional statistical and machine-learning techniques, such as Bayesian regression, logistic regression, linear regression, support vector machine, artificial neural network, deep learning, and reinforcement learning. No seasonal effects exist in cryptocurrency, making it hard to predict using a statistical approach. Traditional statistical methods, although simple to implement and interpret, require a lot of statistical assumptions that could be unrealistic, leaving machine learning as the best technology in this field, being capable of predicting price based on experience. This article provides a comprehensive summary of the previous studies in the field of cryptocurrency price prediction from 2010 to 2020. The discussion presented in this article will help researchers to fill the gap in existing studies and gain more future insight.     

A comparative analysis of cryptocurrency returns and economic policy uncertainty pre- and post-Covid-19

This study analyzes the impact of economic policy uncertainty (EPU) on cryptocurrency returns for a sample of 100 highly capitalized cryptocurrencies from January 2016 to May 2021. The results of the panel data analysis and quantile regression show that increases in global EPU have a positive impact on cryptocurrency returns for lower cryptocurrency returns quantiles and an adverse impact for upper quantiles. In line with the existing literature, the Covid-19 pandemic resulted in higher returns for cryptocurrencies. Inclusion of a Covid-19 dummy in the models strengthened the impact of EPU on cryptocurrency returns. Furthermore, the relationship between the change in EPU and cryptocurrency returns was direct in the pre-Covid-19 period but inverse in the post-Covid-19 period. These results imply that cryptocurrencies act more like traditional financial assets in the post-Covid-19 era.     

Cryptocurrency returns under empirical asset pricing

This study examines the predictability of cryptocurrency returns based on investors' risk premia. Prior studies that have examined the predictability of cryptocurrencies using various economic risk factors have reported mixed results. Our out-of-sample evidence identifies the existence of a significant return predictability of cryptocurrencies based on the cryptocurrency market risk premium. Consistent with capital asset pricing theory (CAPM), our results show that investors often require higher positive returns before taking on any additional risks, particularly in terms of riskier assets like cryptocurrencies. Tests involving the CAPM model demonstrates that the three largest cryptocurrencies have significant exposures to the proposed market factor with insignificant intercepts, demonstrating that the market factor explains average cryptocurrency returns very well.     

Performance-weighted ensembles of random forests for predicting price impact

For any large player in financial markets, the impact of their trading activity represents a substantial proportion of transaction costs. This paper proposes a novel machine learning algorithm for predicting the price impact of order book events. Specifically, we introduce a prediction system based on ensembles of random forests (RFs). The system is trained and tested on depth-of-book data from the BATS and Chi-X exchanges and performance is benchmarked using ensembles of other popular regression algorithms including: linear regression, neural networks and support vector regression. The results show that recency-weighted ensembles of RFs produce over 15% greater prediction accuracy on out-of-sample data, for 5 out of 6 timeframes studied, compared with all benchmarks. Feature importance ranking is used to explore the significance of various market features on the price impact, finding them to be highly variable through time. Finally, a novel procedure for extracting the directional effects of features is proposed and used to explore the features most dominant in the price formation process.     

Investor attention and cryptocurrency: Evidence from wavelet-based quantile Granger causality analysis

This paper investigates the relationship between investor attention and the major cryptocurrency markets by wavelet-based quantile Granger causality. The wavelet analysis illustrates the interdependence between investor attention and the cryptocurrency returns. Multi-scale quantile Granger causality based on wavelet decomposition further demonstrates bidirectional Granger causality between investor attention and the returns of Bitcoin, Ethereum, Ripple and Litecoin for all quantiles, except for the medium. Among them, the Granger causality from investor attention to the returns is relatively very weak for Ethereum. In the short term, the Granger causality from these cryptocurrency returns to investor attention seems symmetric, but in the medium- and long- term, the causality shows some asymmetry. The Granger causality from investor attention to these cryptocurrency returns is asymmetric and varies across cryptocurrencies and time scales. Specifically, investor attention has a relatively stronger impact on the cryptocurrency returns in bearish markets than that in bullish markets in the short term.     

Liquidity in the cryptocurrency market and commonalities across anomalies

We examine how liquidity affects cryptocurrency market efficiency and study commonalities in anomaly performance in cryptocurrency markets. Based on the unique features of cryptocurrencies, we build a model with anonymous traders valuing cryptocurrencies as payments for goods and investment assets, and find that decreases in funding liquidity translate into lower asset liquidity in the cryptocurrency market. Empirically, we observe that many widely recognized stock market anomalies also exist in the cryptocurrency market, although some have opposite long and short legs. We also find evidence that a decrease in cryptocurrency liquidity enhances anomalous returns while preventing the cryptocurrency market from achieving efficiency.     

MAX momentum in cryptocurrency markets

This paper studies the MAX effect, the relationship between maximum daily returns and future returns in the cryptocurrency market. The cryptocurrency market is an ideal setting for the MAX effect due to its lottery-like features (i.e., large positive skewness). Contrary to findings in other markets, we demonstrate that cryptocurrencies with higher maximum daily returns tend to achieve higher returns in the future and call this the “MAX momentum” effect. We also find that the magnitude of the MAX momentum effect varies with market conditions, investor sentiment and the underpricing of cryptocurrencies. Additionally, this effect is robust to longer holding periods, different MAX measures and alternative sample selection criteria.     

Asymmetric interdependencies between large capital cryptocurrency and Gold returns during the COVID-19 pandemic crisis

This article explores asymmetric interdependencies between the twelve largest cryptocurrency and Gold returns, over the period January 2015 – June 2020 within a NARDL (nonlinear autoregressive distributed lag) framework. We focus our analysis on the epicentre of the first wave of the COVID-19 pandemic from March 2020 to June 2020. During this crisis, cryptocurrencies are more correlated and more of them have returns that are cointegrated with Gold returns. Moreover, cryptocurrencies develop a long-term as well as a short-term asymmetric response to Gold returns during the COVID-19 period where most cryptocurrency returns respond more to negative changes and exhibit more persistence with Gold returns. Overall, our most important result confirms that the connectedness between Gold price returns and cryptocurrency returns increase in economic turmoil, such as during the COVID-19 crisis.     

Good volatility,bad volatility,and the cross section of cryptocurrency returns

This paper examines the predictability of realized volatility measures (RVM), especially the realized signed jumps (RSJ), on future volatility and returns. We confirm the existence of volatility persistence and future volatility is more strongly related to the volatility of past positive returns than to that of negative returns in the cryptocurrency market. RSJ-sorted cryptocurrency portfolios yield statistically and economically significant differences in the subsequent portfolio returns. After controlling for cryptocurrency market characteristics and existing risk factors, the differences remain significant. The investor attention explains the predictability of realized jump risk in future cryptocurrency returns.     

Is idiosyncratic volatility priced in cryptocurrency markets?

This paper investigates how idiosyncratic volatility is priced in the cross-section of cryptocurrency returns. By conducting both portfolio-level analysis and Fama-MacBeth regression analysis, we demonstrate that idiosyncratic volatility is positively related to the expected returns of cryptocurrencies. This finding is not subsumed by effects of size, momentum, liquidity, volume, and price and is robust to different weighting schemes, holding periods, and sample sizes. Besides, we find no evidence of temporal relation between idiosyncratic volatility and returns in cryptocurrency markets.     

Forecasting high-frequency excess stock returns via data analytics and machine learning

Borsa Istanbul introduced data analytics to present additional information about its market conditions. We examine whether this product can be utilized via various machine learning methods to predict intraday excess returns. Accordingly, these analytics provide significant prediction ratios above 50% with ideal profit ratios that can reach up to 33%. Among all the methods considered, XGBoost (logistic regression) performs better in predicting excess returns in the long-term analysis (short-term analysis). Results provide evidence for the benefits of both the analytics and the machine learning methods and raise further discussion on the semistrong market efficiency.     

Stock price prediction using DEEP learning algorithm and its comparison with machine learning algorithms

Security indices are the main tools for evaluation of the status of financial markets. Moreover, a main part of the economy of any country is constituted of investment in stock markets. Therefore, investors could maximize the return of investment if it becomes possible to predict the future trend of stock market with appropriate methods. The nonlinearity and nonstationarity of financial series make their prediction complicated. This study seeks to evaluate the prediction power of machine‐learning models in a stock market. The data used in this study include the daily close price data of iShares MSCI United Kingdom exchange‐traded fund from January 2015 to June 2018. The prediction process is done through four models of machine‐learning algorithms. The results indicate that the deep learning method is better in prediction than the other methods, and the support vector regression method is in the next rank with respect to neural network and random forest methods with less error.     

How well do investor sentiment and ensemble learning predict Bitcoin prices?

Investor sentiment is widely recognized as the major determinant of cryptocurrency prices. Although earlier research has revealed the influence of investor sentiment on cryptocurrency prices, it has not yet generated cohesive empirical findings on an important question: How effective is investor sentiment in predicting cryptocurrency prices? To address this gap, we propose a novel prediction model based on the Bitcoin Misery Index, using trading data for cryptocurrency rather than judgments from individuals who are not Bitcoin investors, as well as bagged support vector regression (BSVR), to forecast Bitcoin prices. The empirical analysis is performed for the period between March 2018 and May 2022. The results of this study suggest that the addition of the sentiment index enhances the predictive performance of BSVR significantly. Moreover, the proposed prediction system, enhanced with an automatic feature selection component, outperforms state-of-the-art methods for predicting cryptocurrency for the next 30 days.     

Non-linear relationship between oil and cryptocurrencies: Evidence from returns and shocks

The price instabilities between oil prices and cryptocurrencies have motivated the current study to examine the nonlinear relationship between oil returns/shocks and cryptocurrencies during March 3, 2018 to October 10, 2021. We employed a novel methodology of cross-quantilogram to unveil the nonlinearity and asymmetry between oil shocks and cryptocurrencies. We find that when markets are normal and bullish, there is a positive correlation between oil returns and cryptocurrency returns at first lag; however, there is a negative correlation between oil returns and cryptocurrencies in all market conditions. Moreover, rising fluctuations in oil demand shocks brings significant movement in cryptocurrency returns in bearish market conditions and it is unlikely that oil demand shocks and cryptocurrencies returns move in same directions. Given these results, we proposed useful implications for policymakers, strategists, regulators, financial market participants, and investors to hedge/diversify their risk.     

On the topology of cryptocurrency markets

Cryptocurrency markets are characterised by high volatility, high returns and comparative immaturity relative to equity and commodity markets. Topological Data Analysis (TDA) persistence norms are effective tools for the analysis of noisy dynamical systems like the cryptocurrency markets. We show how information from the shape of daily return data adds additional inference on activity within the cryptocurrency markets. TDA persistence norms embed volatility and connectedness between coins as well as incorporating information from uncertainty indexes, financial market performance and commodity returns. Our TDA measures are robust to noise and are consistent across a raft of alternative coin selections. Further, we exposit how persistence norms peak to forewarn of crashes and stay low as markets face exogenous shocks. We demonstrate the clear advantages of TDA for the study of cryptocurrency markets and develop the next steps for exploiting the potential of TDA for application to cryptocurrency markets.     

Data vs. information: Using clustering techniques to enhance stock returns forecasting

This paper explores the use of clustering models of stocks to improve both (a) the prediction of stock prices and (b) the returns of trading algorithms.We cluster stocks using k-means and several alternative distance metrics, using as features quarterly financial ratios, prices and daily returns. Then, for each cluster, we train ARIMA and LSTM forecasting models to predict the daily price of each stock in the cluster. Finally, we employ the clustering-empowered forecasting models to analyze the returns of different trading algorithms.We obtain three key results: (i) LSTM models outperform ARIMA and benchmark models, obtaining positive investment returns in several scenarios; (ii) forecasting is improved by using the additional information provided by the clustering methods, therefore selecting relevant data is an important preprocessing task in the forecasting process; (iii) using information from the whole sample of stocks deteriorates the forecasting ability of LSTM models.These results have been validated using data of 240 companies of the Russell 3000 index spanning 2017 to 2022, training and testing with different subperiods.     

Nonlinear tail dependence in cryptocurrency-stock market returns: The role of Bitcoin futures

We investigate the median and tail dependence between cryptocurrency and stock market returns of BRICS and Developed countries using a newly developed nonparametric cumulative measure of dependence over the period January 4, 2016 – December 31, 2019 as well as before and after the introduction of Bitcoin futures on December 17, 2017. The new measure is model-free and permits measuring tail risk. The results highlight the leading role of S&P500, Nasdaq and DAX 30 in predicting BRICS and developed countries’ stock market returns. Among BRICS countries, BVSP shows a starring role in predicting stock market returns. BSE 30 is the most predictor of cryptocurrencies, which have a little predictability on stock market returns. Ethereum has the leading role in predicting cryptocurrencies and stock market returns followed by Bitcoin. Tail dependence shows substantial role of S&P500, Nasdaq and BVSP in predicting stock market returns. Subsample analysis show the role of Bitcoin futures in reshaping the mean and tail dependence between cryptocurrency and stock market returns. Our results have important policy implications for portfolio managers, hedge funds and investors.     

Time-varying neural network for stock return prediction

We consider the problem of neural network training in a time-varying context. Machine learning algorithms have excelled in problems that do not change over time. However, problems encountered in financial markets are often time varying. We propose the online early stopping algorithm and show that a neural network trained using this algorithm can track a function changing with unknown dynamics. We compare the proposed algorithm to current approaches on predicting monthly US stock returns and show its superiority. We also show that prominent factors (such as the size and momentum effects) and industry indicators exhibit time-varying predictive power on stock returns. We find that during market distress, industry indicators experience an increase in importance at the expense of firm level features. This indicates that industries play a role in explaining stock returns during periods of heightened risk.     

Equilibrium Bitcoin Pricing

We offer a general equilibrium analysis of cryptocurrency pricing. The fundamental value of the cryptocurrency is its stream of net transactional benefits, which depend on its future prices. This implies that, in addition to fundamentals, equilibrium prices reflect sunspots. This in turn implies multiple equilibria and extrinsic volatility, that is, cryptocurrency prices fluctuate even when fundamentals are constant. To match our model to the data, we construct indices measuring the net transactional benefits of Bitcoin. In our calibration, part of the variations in Bitcoin returns reflects changes in net transactional benefits, but a larger share reflects extrinsic volatility.