ANALISIS PERBANDINGAN ALGORITMA MULTI-OBJEKTIF UNTUK OPTIMISASI PORTOFOLIO SAHAM: STUDI MOPSO (CROWDING DISTANCE, GRID-BASED) DAN NSGA-II BERDASARKAN KLASTER K-MEANS++ DAN SPECTRAL

Maulana Fajar Aji Prasetya, Prof. Dr. Abdurakhman, S.Si., M.Si.

2025 | Skripsi | STATISTIKA

Keterbatasan model optimisasi portofolio klasik seperti Mean Variance dalam menghadapi volatilitas pasar modal mendorong kebutuhan akan strategi yang lebih adaptif. Penelitian ini bertujuan mengembangkan pendekatan komprehensif untuk optimisasi portofolio saham multi-objektif dengan menggabungkan analisis klaster dan algoritma metaheuristik. Data saham dari indeks LQ45 periode Mei 2024 hingga Mei 2025 digunakan sebagai studi kasus. Saham-saham tersebut dikelompokkan menggunakan K-Means++ dan Spectral Clustering, yang masing-masing mengidentifikasi 5 dan 6 klaster optimal. Proses optimisasi melibatkan MO-MV sebagai baseline, serta MOPSO varian Crowding Distance dan Grid-Based, dan NSGA-II. Hasil penelitian menunjukkan bahwa MOPSO, khususnya varian Crowding Distance dan Grids dengan klasterisasi Spectral Clustering, mencapai sharpe ratio tertinggi sebesar 1.320, mengungguli MO-MV. Namun, NSGA-II secara konsisten menghasilkan sharpe ratio negatif, mencerminkan perbedaan dalam strategi pencarian di kondisi pasar yang menantang. NSGA-II cenderung mempertahankan keragaman luas sementara MOPSO lebih efektif mengeksploitasi solusi efisien yang sangat spesifik.

The limitations of classical portfolio optimization models such as Mean Variance in dealing with stock market volatility have driven the need for more adaptive strategies. This study aims to develop a comprehensive approach to multi-objective stock portfolio optimization by combining cluster analysis and metaheuristic algorithms. Stock data from the LQ45 index for the period May 2024 to May 2025 was used as a case study. The stocks were grouped using K-Means++ and Spectral Clustering, which identified 5 and 6 optimal clusters, respectively. The optimization process involves MO-MV as the baseline, as well as MOPSO Crowding Distance and Grid-Based variants, and NSGA-II. The results showed that MOPSO, particularly the Crowding Distance and Grids variants with Spectral Clustering clustering, achieved the highest sharpe ratio of 1.320, outperforming MO-MV. However, NSGA-II consistently generated a negative sharpe ratio, reflecting differences in search strategies under challenging market conditions. NSGA-II tends to maintain broad diversity while MOPSO more effectively exploits highly specific efficient solutions. 

Kata Kunci : optimisasi portofolio, MOPSO, NSGA-II, K-Means++, Spectral, saham.