Cryptocurrency Returns Clustering Using Japanese Candlesticks: Towards a Programmatic Trading System

Abstract

Cryptocurrency trading drove the attention of individual traders during and after the lockdown period caused by COVID-19 restrictions. Trading systems use Japanese candlesticks-derived technical indicators to decide on behalf of traders. They offer insights into the market trend and help traders to decide how to manage their cryptocurrency portfolio. Japanese candlesticks help visualize the movement of cryptocurrencies’ prices over a given period. This transformation is widely used to forecast the future trend, volatility, and prices of a cryptocurrency. Most of the research on forecasting returns suggests using three classes, uptrend, downtrend, and insignificant changes. Within this paper, we applied KMeans clustering to the Japanese candlesticks over a daily period of five of the highest capitalized non-stable coins (Bitcoin, Ethereum, BNB coin, Cardano, and Solana). Results indicates that the optimal number of clusters is ranging from 5 to 6 clusters using the elbow method for all the considered cryptocurrencies. The range of obtained results suggests that we should opt for a per cryptocurrency number of classes when dealing with cryptocurrencies classification tasks rather than using the three classes mentioned above.