An Identification and Estimation of Stock Price Pattern Equations using K-Means

Clustering is a method of grouping similar objects together. Numerous clustering methods have been invented to date. However, with the advent of machine learning technologies and with the availability of large scale datasets, clustering has been revived. This paper applies K-means clustering to study price patterns in a large scale dataset of stock prices. It is novel in that it, among other things, reshapes the large scale complex dataset into a sequence of chained collocations of stock prices, which are then clustered at these collocation points. The approach managed to extract several stylized price patterns by a diverse set of clusters of different sizes and shapes, exhibiting linear independence and power-law distribution in size. Empirical estimations of these and other features support the validity and robustness of the clustering methodology used in this paper. Altogether, the paper makes a strong case in favor of the K-means clustering for the study of stock-price patterns.