Structured PREreview of A Comprehensive Survey of Cryptocurrency Forecasting: Methods, Trends, and Challenges

Does the introduction explain the objective of the research presented in the preprint?

Yes

The introduction clearly explains that the objective of the research is to provide a comprehensive survey of cryptocurrency forecasting, tracing its evolution and significant growth in the global financial arena. Specifically, the survey aims to consolidate existing research by categorizing and analyzing 234 scholarly articles across various methodologies including machine learning, deep learning, deep reinforcement learning, and statistical methods. By synthesizing a wide array of literature, the paper intends to furnish a comprehensive overview of prevailing trends, methodologies, and challenges in the domain of cryptocurrency forecasting, serving as a valuable resource for researchers and investors navigating this dynamic field.

Are the methods well-suited for this research?

Highly appropriate

The methods employed, which center on a meticulous review, categorization, and analysis of 234 scholarly articles, are exceptionally well-suited for achieving the research objective of producing a comprehensive survey of cryptocurrency forecasting. This systematic approach allows the authors to consolidate existing research by organizing it into distinct methodological categories including machine learning, deep learning, deep reinforcement learning, and statistical methodologies. Analyzing key trends such as yearly publication rates, methodological distributions, input features, training/testing splits, and forecasting time horizons directly fulfills the aim of providing a detailed overview of prevailing practices and challenges. Lastly, incorporating detailed case studies, such as examining performance disparities between backtesting and forward testing, underscores the practical challenges inherent in the field, arming researchers and investors with a nuanced understanding of cryptocurrency forecasting.

Are the conclusions supported by the data?

Highly supported

The conclusions of the survey are strongly supported by the extensive data analysis presented throughout the preprint, beginning with the consolidation and categorization of 234 scholarly articles across machine learning, deep learning, deep reinforcement learning, and statistical methodologies. The comprehensive nature of the survey is evident in findings that detail yearly publication rates, methodological distributions, input features, and time horizons. Specific data points show Deep Learning is the most frequently used learner type (50.3%), with Long Short Term Memory (LSTM) being the predominant individual methodology utilized in 109 publications. Furthermore, the analysis reveals that research overwhelmingly focuses on the 24 hour time horizon (67.5%), and the most frequently used metrics are Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Mean Absolute Percentage Error (MAPE). The conclusion regarding the practical challenges of implementing forecasting strategies in real world scenarios is substantiated by the case study "Examining the performance differences between backtesting and forward testing". This study illustrated a divergence in performance, noting that while algorithms showed notable proficiency in cumulative profit during backtesting (mid 2022 to mid 2023), some struggled to sustain those gains during the forward testing period (last six months of 2023). The inclusion of a "Social Data Exploration in Cryptocurrency Trends" case study further validates the conclusion that external factors influence the market by providing visual evidence correlating Bitcoin prices with Google search trends, Reddit comments, and cryptocurrency news articles.

Are the data presentations, including visualizations, well-suited to represent the data?

Highly appropriate and clear

The data presentations, including visualizations, are highly well-suited to represent the data, comprehensively illustrating the findings derived from the analysis of 234 scholarly articles on cryptocurrency forecasting. The use of both bar charts and pie charts effectively visualizes complex data distributions, enhancing clarity and understanding across multiple facets of the research. For instance, bar charts vividly portray yearly publication trends, showing a significant upswing in research activity in 2022 and 2023, while also detailing the frequency of specific methodologies like Long Short-Term Memory, the most utilized method among surveyed papers. Pie charts complement this by providing proportional representations, clearly illustrating that 67.5% of published research papers focused on the 24-hour time horizon and that Deep Learning algorithms constitute the largest segment of learner types at 50.3%. Additionally, line charts are employed to trace the consistent prominence of key evaluation metrics such as Root Mean Square Error, Mean Absolute Error, and Mean Absolute Percentage Error over successive years, offering valuable insights into enduring research practices. The visual data presentations also effectively support the case studies, with line charts illustrating the correlation between Bitcoin prices and external factors like Google search trends, Reddit comments, and cryptocurrency news articles, which is crucial for understanding market behavior and sentiment. The use of a word cloud further aggregates and visually highlights the prevalence of key forecasting models such as LSTM, ARIMA, and SVM, providing an accessible overview of terminology in the cryptocurrency forecasting domain. Scatter plots are utilized to show the distribution of training and testing data samples, indicating that most studies used between 1000 and 3000 data samples for model evaluation. These diverse and integrated visualizations ensure that the survey’s extensive findings regarding methodologies, trends, and challenges are clearly and effectively communicated to a broad audience of researchers and investors.

How clearly do the authors discuss, explain, and interpret their findings and potential next steps for the research?

Very clearly

The authors clearly discuss, explain, and interpret their findings through an extensive analysis of 234 scholarly articles, presenting key trends across various methodologies and market dynamics. Interpretation is facilitated by dedicated sections and case studies which illuminate trends like the dominance of Deep Learning models, particularly Long Short Term Memory, which is the most frequently utilized individual method in the surveyed papers. The research interprets the strong correlation between Bitcoin prices and external social data factors, noting the surge in searches corresponding to the 2021 price high and increased social media engagement during significant market events like the FTX collapse. A specific case study explains the practical challenges investors face by illustrating performance disparities, noting that algorithms showing proficiency in historical backtesting often struggled to sustain gains during real world forward testing. The paper comprehensively addresses potential next steps by detailing numerous challenges and open problems that define the future research agenda. These include fundamental issues such as models overfitting, survivorship bias, data quality limitations, a lack of model interpretability, and the difficulty of quantifying risk and uncertainty, with authors suggesting routes for future research to tackle these critical issues, such as employing probabilistic methods like Bayesian inference.

Is the preprint likely to advance academic knowledge?

Highly likely

The preprint is highly likely to advance academic knowledge because it offers significant contributions that substantially consolidate and expand the understanding of cryptocurrency forecasting. This survey conducted an extensive review of 234 scholarly articles, meticulously categorizing them across diverse methodologies including machine learning, deep learning, deep reinforcement learning, and statistical models. The analysis provides detailed insights into prevailing trends, such as yearly publication rates, methodological distributions, input features, and forecasting time horizons. Furthermore, the study contributes unique practical knowledge by investigating crucial issues through dedicated case studies, specifically examining the performance disparities between backtesting and forward testing, and exploring the impact of social media data on cryptocurrency prices. By synthesizing this vast body of literature and clearly outlining numerous challenges and open problems, the paper serves as a valuable resource that offers researchers and investors with a solid foundation for navigating this complex and evolving field.

Would it benefit from language editing?

No

The language used throughout the preprint is professional and clear

Would you recommend this preprint to others?

Yes, it’s of high quality

The survey methodologically consolidates existing research by meticulously categorizing findings across machine learning, deep learning, deep reinforcement learning, and statistical methodologies. The analysis provides key insights, such as Deep Learning being the most widely used learner type, and identifies specific dominant methods, such as Long Short-Term Memory. More importantly, the paper includes practical case studies that highlight the challenges investors face, specifically examining the significant performance disparities between backtesting and forward testing strategies and exploring how external factors, such as social media data, impact market trends. Finally, the authors clearly articulate open problems and challenges, setting a robust agenda for future research on addressing model overfitting and data quality limitations.

Is it ready for attention from an editor, publisher or broader audience?

Yes, as it is

Competing interests

The author declares that they have no competing interests.

Use of Artificial Intelligence (AI)

The author declares that they did not use generative AI to come up with new ideas for their review.