Pembentukan Portofolio Saham Menggunakan Klastering Time Series K-Medoid dengan Ukuran Jarak Dynamic Time Warping

Abstract

This research will present the formation of stock portfolios by preprocessing data using time series clustering with a distance measure of Dynamic Time Warping (DTW). First, stocks are grouped into several clusters using the Partitioning Around Medoids (PAM) time series cluster based on the DTW distance measure. After the clustering process, stocks are selected to represent each cluster to build the optimum portfolio. The stock selected from each cluster is the one with the highest Sharpe ratio. The optimal portfolio is determined using three portfolio models, namely: the classic MV portfolio model, the FMCD robust MV portfolio model and the S robust portfolio model. Using this procedure, an optimum portfolio can be obtained efficiently if there are many stocks involved in the portfolio formation process. Sharpe ratio is used to measure the performance of the portfolios. The results of the empirical study show that the portfolio performance generated using the PAM time series clustering with DWT distance dissimilarity measure combined with the classic MV portfolio model outperforms the resulting portfolio performance in combination with other models.