Robust Mean–Variance Portfolio Selection Using Cluster Analysis: A Comparison between Kamila and Weighted K-Mean Clustering

Abstract

This study presents robust portfolio selection using cluster analysis of mixed-type data. For this empirical study, the daily price data of LQ45 index stocks listed on the Indonesia Stock Exchange were employed. First, six stocks clusters are formed by using the KAMILA algorithm on a combination of continuous and categorical variables. For comparison purposes, weighted k-means cluster analysis was also undertaken. Second, stocks that were representative of each cluster, those with the highest Sharpe ratios, were selected to create a portfolio. The optimum portfolio was determined through classic (non-robust) and the robust estimation methods of fast minimum covariance determinant (FMCD) and S estimation. Using a robust procedure enables the best-performing portfolio to be created efficiently when selecting assets from a large number of stocks, especially as the results are largely unaffected in the presence of outliers. This study found that the performance of the portfolio developed with the KAMILA clustering algorithm and robust FMCD estimation outperformed those created by other methods.