Application of Agglomerative Hierarchical Clustering Based on Rolling Horizon and Minimum Variance in IDX30 Stock Portfolio Optimization

Khoirunnisa Puspa Negari, E. Setiawan

Abstract

Portfolio diversification is a strategy to reduce investment risk by spreading assets across stocks with different movement patterns. However, most investors tend to select stocks based on highest historical returns and exhibit herding behavior without considering inter-stock correlations, risking undiversified portfolios. Stock movement patterns are dynamic, requiring methods that accommodate such changes. This study aims to cluster IDX30 stocks using agglomerative hierarchical clustering based on rolling horizon to identify stocks with relatively consistent movement patterns, compare diversification characteristics, and evaluate portfolio performance using minimum variance and equal weight. Data consists of daily closing prices of IDX30 stocks from January 3, 2022 to December 30, 2024 obtained from Yahoo Finance. Results show stocks tend to cluster based on sectoral similarity with cluster membership proportion of  for  and  for . Horizons 5 and 12 were selected as portfolio construction periods due to having the highest number of consistent stocks. Clustering portfolios exhibit better diversification with more low-correlated stock pairs compared to non-clustering portfolios. Minimum variance optimization shows the  clustering portfolio outperforms with Sharpe ratios of  at horizon 5 and  at horizon 12 compared to  and non-clustering portfolios. Conversely, equal weight shows non-clustering portfolios outperform clustering portfolios. During the evaluation period, the  clustering portfolio with minimum variance achieved average returns of  for 30 days and  for 60 days after horizon 5 and demonstrated relatively stable performance under extreme market conditions at horizon 12.

Source: semanticscholar · PDF

Read the AI summary, key takeaways and discussion on WOBR Quant Research.