A new method for similarity and anomaly detection in cryptocurrency markets

We propose a new approach using the MJ1 semi-metric, from the more general MJp class of semi-metrics [1], to detect similarity and anomalies in collections of cryptocurrencies. Since change points are signals of potential risk, we apply this metric to measure distance between change point sets, with respect to returns and variance. Such change point sets can be identified using algorithms such as the Mann-Whitney test, while the distance matrix is analysed using three approaches to detect similarity and identify clusters of similar cryptocurrencies. This aims to avoid constructing portfolios with highly similar behaviours, reducing total portfolio risk.