Index Tracking Based on Dynamic Time Warping and Constrained k-medoids Clustering

Abstract

Index tracking is a passive investment strategy by replicating a financial market index using its constituents. In this paper, index tracking is addressed based on $k-$medoids clustering. $k-$medoids clustering is formulated as a valuation-constrained $k-$median problem to cluster index constituents. The dissimilarity coefficients among stocks are measured by using dynamic time warping. Experimental results of index tracking on four major indices are elaborated to demonstrate that the tracking performance of the proposed method with dynamic time warping is superior to that with Pearson correlation coefficients.