Enhancing Pairs Trading Strategies in the Cryptocurrency Industry using Machine Learning Clustering Algorithms

Abstract

Conventional pair trading methods, which rely on statistical and linear assumptions, often struggle to cope with the high volatility and dynamic nature of cryptocurrency markets. This study explores how pair trading strategies might be improved by using machine learning clustering algorithms to uncover latent links between cryptocurrencies. Specifically, it employs unsupervised clustering techniques k-means, hierarchical clustering, and affinity propagation on daily closing prices of the top 50 cryptocurrencies, selected based on their market capitalization and daily trading volume, from January 2021 to November 2024. The methodology includes data preprocessing, exploratory data analysis, clustering, and cointegration tests for pair selection. The main findings show that clustering algorithms can efficiently group cryptocurrencies based on similar behavioural price patterns, with affinity propagation outperforming other models in cluster definition. The study reveals 21 pairs with strong cointegration strategies among the chosen cryptocurrencies, indicating their appropriateness for trading. The study highlights the effectiveness of clustering algorithms in tackling cryptocurrency market volatility, optimizing pair selection, and adapting to dynamic conditions. It emphasizes the transformative potential of machine learning in enhancing trading techniques and efficiency in the cryptocurrencies market. The practical implications include advancing trading strategies for cryptocurrencies investors by incorporating machine learning algorithms to enhance market efficiency and profitability