Efficient Stock Portfolio Construction by Means of Clustering

When investors start to use statistical methods to optimise their stock market investment decisions, one of fundamental problems is constructing a well‑diversified portfolio consisting of a moderate number of positions. Among a multitude of methods applied to the task, there is a group based on dividing all companies into a couple of homogeneous groups followed by picking out a representative from each group to create the final portfolio. The division stage does not have to coincide with the sector affiliation of companies. When the division is performed by means of clustering of companies, a vital part of the process is to establish a good number of clusters. The aim of this article is to present a novel technique of portfolio construction based on establishing a numer of portfolio positions as well as choosing cluster representatives. The grouping methods used in the clustering process are the classical k‑means and the PAM (Partitioning Around Medoids) algorithm. The technique is tested on data concerning the 85 biggest companies from the Warsaw Stock Exchange for the years 2011–2016. The results are satisfactory with respect to the overall possibility of creating a clustering‑based algorithm requiring almost no intervention on the part of the investor.