Integrating Density-Based Clustering and Sentiment Analysis for Portfolio Optimization: An Empirical Study on IDX ESG Leaders

Abstract

The heightened uncertainty in Indonesia’s capital market in early 2025, driven by capital outflows and macroeconomic instability, underscores the need for more robust data-driven portfolio construction frameworks beyond conventional fundamental analysis. This study develops an integrated statistical learning framework that combines transformer-based sentiment analysis and density-based clustering to improve portfolio optimization for stocks listed in the IDX ESG Leaders index. Market sentiment is quantified from financial news using the FinBERT model, producing structured sentiment scores that are incorporated as exogenous inputs. Subsequently, the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm is applied to identify latent structures within the asset space based on sentiment similarity, enabling non-parametric cluster formation without imposing distributional assumptions. The empirical findings indicate that IDX ESG Leaders stocks are characterized by predominantly positive sentiment distributions, with clustering results revealing two principal clusters and one noise component. To assess portfolio performance, the resulting clusters are integrated into a mean–variance optimization framework. The results demonstrate that portfolios constructed using sentiment-informed clustering achieve superior risk-adjusted performance, reflecting enhanced diversification and reduced estimation bias relative to conventional approaches. In contrast, portfolios exhibiting high concentration or sentiment homogeneity tend to yield lower efficiency, despite strong fundamental characteristics. Overall, this study demonstrates the statistical value of incorporating unstructured textual information and unsupervised learning techniques into portfolio optimization, providing empirical evidence that supports the integration of sentiment-driven features in data-driven asset allocation strategies.