Time series clustering for high-dimensional portfolio selection: a comparative study

In high-dimensional portfolio selection, traditional asset allocation techniques often yield suboptimal results out-of-sample, while equally weighted portfolios have shown better performances in such scenarios. To leverage the advantages of diversification while addressing the curse of dimensionality, we turn to clustering techniques. Specifically, we explore the application of k-meansclustering for time series, which offers a clear financial interpretation as the prototype of each cluster represents an equally weighted portfolio of the assets within the cluster. In this paper, we conduct a comprehensive comparison of various time series clustering techniques in the context of portfolio performance. By evaluating the out-of-sample performance of portfolios constructed using different clustering approaches, we aim to identify the most effective method for investment purposes.