ADS Capstone Chronicles Revised

3.3 Kristoufek (2013)

environments. Additionally, the study relies on historical data without testing the models in real time market scenarios, which could be crucial for understanding their practical applicability in dynamic trading contexts.

Kristoufek used Google Trends and Wikipedia data to provide an analytical perspective on the interplay between Bitcoin prices and online search activities. Through rigorous econometric analysis, the author identified a bidirectional relationship between Bitcoin prices and search volumes, suggesting while increasing prices spur public interest, heightened online activity can also predict subsequent price movements. This research significantly contributes to understanding the feedback loops in digital currency markets; however, it predominantly focuses on observational data without incorporating the potential for these relationships to be exploited in predictive modeling. Moreover, while the study offers valuable insights into market sentiment and behavior, it needs to explore how traders could practically apply these factors for real-time decision-making in cryptocurrency trading. McNally et al. ’ s pivotal focus is exploring machine learning techniques to predict Bitcoin prices, specifically employing neural networks and linear regression models. The authors methodically applied these techniques to a historical Bitcoin price data dataset to ascertain predictive accuracy. Their results notably demonstrated machine learning models can outperform traditional time-series forecasting methods, especially in capturing the nonlinear patterns often observed in cryptocurrency markets. However, the study has its limitations. One notable gap is examining model performance across diverse market conditions or cryptocurrencies other than Bitcoin. This raises questions about the generalizability of the findings to other volatile cryptocurrency 3.4 McNally et al. (2018)

3.5 Sattarov et al. (2020 )

Sattarov et al.’s main theme is the application of various trading strategies for cryptocurrencies. These strategies include the crossover strategy, day trading, swing trading, scalping, and positional trading. The authors discuss how these strategies can optimize profits but highlight the time and expertise required to understand them fully and how they might only sometimes yield the best results. Machine learning is introduced to help traders and investors achieve higher profits when applying the correct algorithm. The article focuses on deep reinforcement learning (DRL), which can predict when to buy, sell, or hold. The authors compare traditional trading strategies with their DRL model, measuring success through profit percentage metrics. They used historical data from Bitcoin, Ethereum, and Litecoin to train their model and conducted controlled experiments to compare the old techniques with their new modeling approach. Their findings indicate traditional trading methods had lower profit percentages and were less effective in volatile markets like cryptocurrency. With its unique adaptability, the DRL model they developed can learn and adapt to market fluctuations, thereby making better trading decisions and providing reassurance in volatile market conditions. However, it ’ s essential to acknowledge gaps in the research. One significant gap is the need for more evidence of the model ’ s performance in real-time trading. While the model is proven conceptually using historical data, its

71