Title: Community Detection in Cryptocurrencies with Potential Applications to Portfolio Diversification

Abstract: In this paper, the cross-correlations of cryptocurrency returns are analysed. The paper examines one years worth of data for 146 cryptocurrencies from the period January 1 2019 to December 31 2019. The cross-correlations of these returns are firstly analysed by comparing eigenvalues and eigenvector components of the cross-correlation matrix C with Random Matrix Theory (RMT) assumptions. Results show that C deviates from these assumptions indicating that C contains genuine information about the correlations between the different cryptocurrencies. From here, Louvain community detection method is applied as a clustering mechanism and 15 community groupings are detected. Finally, PCA is completed on the standardised returns of each of these clusters to create a portfolio of cryptocurrencies for investment. This method selects a portfolio which contains a number of high value coins when compared back against their market ranking in the same year. In the interest of assessing continuity of the initial results, the method is also applied to a smaller dataset of the top 50 cryptocurrencies across three time periods of T = 125 days, which produces similar results. The results obtained in this paper show that these methods could be useful for constructing a portfolio of optimally performing cryptocurrencies.