Penggerombolan Indeks LQ45 Berdasarkan Pola Harga Menggunakan Jarak Dynamic Time Warping dan Global Alignment Kernel

Abstract

Krisis perang dagang antara Amerika Serikat dan Cina pada awal 2025 menyebabkan tekanan terhadap pasar saham, termasuk indeks LQ45 di Indonesia. Penelitian ini bertujuan untuk menggerombolkan saham-saham dalam indeks LQ45 berdasarkan pola pergerakan harga selama periode krisis menggunakan pendekatan shape-based time series clustering. Proses penggerombolan dilakukan dengan metode hierarki menggunakan tiga jenis pautan (rataan, tunggal, dan lengkap) dan dua ukuran ketidakmiripan, yaitu Dynamic Time Warping (DTW) dan Global Alignment Kernel (GAK). Evaluasi model dilakukan menggunakan metrik internal berupa Silhouette, Calinski-Harabasz Index (CHI), dan Dunn Index. Hasil menunjukkan bahwa kombinasi pautan rataan dan jarak GAK menghasilkan performa terbaik dengan empat gerombol optimum. Model ini memperoleh nilai Silhouette sebesar 0,447, CHI sebesar 42,46, dan Dunn Index sebesar 1,914, yang mengindikasikan kualitas penggerombolan yang baik. Tiap gerombol menunjukkan pola harga dan komposisi sektoral yang berbeda. Salah satu gerombol yang didominasi sektor Financials, Energy, dan Infrastructures menunjukkan pemulihan harga signifikan, sedangkan gerombol lainnya cenderung menunjukkan pemulihan yang lemah atau bahkan tidak terpengaruh penurunan tajam saat trading halt.

The trade war crisis between the United States and China in early 2025 exerted pressure on the stock market, including the LQ45 index in Indonesia. This study aims to cluster stocks in the LQ45 index based on price movement patterns during the crisis period using a shape-based time series clustering approach. The clustering process was conducted using hierarchical methods with three linkage types (average, single, and complete) and two dissimilarity measures, namely Dynamic Time Warping (DTW) and Global Alignment Kernel (GAK). Model evaluation was carried out using internal metrics including Silhouette, Calinski-Harabasz Index (CHI), and Dunn Index. The results show that the combination of average linkage and GAK distance produced the best performance with four optimal clusters. This model achieved a Silhouette Score of 0.447, a CHI value of 42.46, and a Dunn Index of 1.914, indicating good clustering quality. Each cluster demonstrated distinct price patterns and sectoral compositions. One cluster, dominated by the Financials, Energy, and Infrastructures sectors, showed significant price recovery, while the other clusters tended to have low price recovery or even were unaffected by the sharp declines during the trading halt periods.