Analisis Peramalan Saham Menggunakan Model ARIMA-GARCH dengan Estimasi Risiko VaR Berdasarkan Hasil Clustering K-Medoids Berbasis DTW

Yesinta, Yesinta (2026) Analisis Peramalan Saham Menggunakan Model ARIMA-GARCH dengan Estimasi Risiko VaR Berdasarkan Hasil Clustering K-Medoids Berbasis DTW. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5006221028-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Investasi saham selalu dihadapkan pada risiko yang berasal dari fluktuasi return dan volatilitas harga, sehingga diperlukan metode yang mampu mengelompokkan saham berdasarkan kesamaan pola pergerakan return sekaligus mengukur tingkat risikonya. Penelitian ini bertujuan untuk mengelompokkan saham pada indeks IDX ESG Leaders menggunakan metode K-Medoids berbasis Dynamic Time Warping (DTW), menentukan model ARIMA-GARCH terbaik pada saham prototype setiap klaster serta mengevaluasi kinerjanya ketika diterapkan pada seluruh anggota klaster, dan mengestimasi risiko investasi menggunakan Value at Risk (VaR). Data yang digunakan berupa return harian 14 saham konstituen indeks IDX ESG Leaders periode 5 Januari 2021–30 Desember 2025 dengan pembagian data training dan testing sebesar 80:20. Hasil penelitian menunjukkan bahwa jumlah klaster optimal adalah dua klaster, didukung oleh nilai Silhouette Index sebesar 0,219207 dan Pseudo-F sebesar 4,1252. Klaster 1 terdiri atas 9 saham dengan UNVR sebagai medoid, sedangkan Klaster 2 terdiri atas 5 saham dengan AKRA sebagai medoid. Model terbaik yang diperoleh adalah ARIMA(0,0,0)-GARCH(1,1) untuk Prototype Klaster 1 dan ARIMA(2,0,1)-GARCH(1,1) untuk Prototype Klaster 2. Evaluasi pada data testing menggunakan MSE, MAE, dan RMSE menunjukkan bahwa model Prototype Klaster 1 menghasilkan nilai masing-masing sebesar 0,00125507, 0,02424975, dan 0,03542701, sedangkan Prototype Klaster 2 menghasilkan nilai masing-masing sebesar 0,00090125, 0,01974911, dan 0,03002079. Penerapan struktur model pada seluruh anggota klaster juga menunjukkan performa yang konsisten, dengan nilai MSE, MAE, dan RMSE data testing berturut-turut berkisar 0,00032738–0,00101813, 0,01334153–0,02166421, dan 0,01809356–0,03190822 pada Klaster 1, serta 0,00046369–0,00087399, 0,01439970–0,02007652, dan 0,02153338–0,02956331 pada Klaster 2. Hasil peramalan return selama 10 hari menunjukkan pola yang relatif stabil. Estimasi Value at Risk (VaR) pada tingkat kepercayaan 90%, 95%, dan 99% menunjukkan bahwa pada tingkat kepercayaan 99%, potensi kerugian terbesar terdapat pada EXCL sebesar −0,107764 untuk Klaster 1 dan TOWR sebesar −0,098700 untuk Klaster 2. Secara keseluruhan, Klaster 1 memiliki variasi tingkat risiko yang lebih tinggi dibandingkan Klaster 2 yang relatif lebih homogen. Hasil penelitian ini diharapkan dapat menjadi referensi dalam analisis risiko investasi serta sebagai bahan pertimbangan bagi investor dalam pengambilan keputusan investasi dan pengelolaan portofolio saham berbasis ESG.
=========================================================================================================================================
Stock investment is inherently exposed to risks arising from fluctuations in returns and price volatility. Therefore, an approach capable of clustering stocks based on similarities in return movement patterns while quantitatively measuring investment risk is required. This study aims to cluster stocks listed in the IDX ESG Leaders Index using the Dynamic Time Warping (DTW)-based K-Medoids method, determine the best ARIMA-GARCH model for the prototype (medoid) of each cluster and evaluate its performance when applied to all cluster members, and estimate investment risk using Value at Risk (VaR). The data used consist of daily returns of 14 constituent stocks of the IDX ESG Leaders Index covering the period from 5 January 2021 to 30 December 2025, with an 80:20 training-testing data split. The results indicate that the optimal clustering solution consists of two clusters, supported by the highest Silhouette Index of 0.219207 and Pseudo-F value of 4.1252. Cluster 1 consists of 9 stocks with UNVR as the medoid, while Cluster 2 consists of 5 stocks with AKRA as the medoid. The best models obtained were ARIMA(0,0,0)-GARCH(1,1) for the Cluster 1 prototype and ARIMA(2,0,1)-GARCH(1,1) for the Cluster 2 prototype. Evaluation on the testing data using MSE, MAE, and RMSE showed that the Cluster 1 prototype model produced values of 0.00125507, 0.02424975, and 0.03542701, respectively, while the Cluster 2 prototype model produced values of 0.00090125, 0.01974911, and 0.03002079, respectively. Applying the model structure to all cluster members also demonstrated consistent performance, with testing MSE, MAE, and RMSE values ranging from 0.00032738–0.00101813, 0.01334153–0.02166421, and 0.01809356–0.03190822, respectively, for Cluster 1, and from 0.00046369–0.00087399, 0.01439970–0.02007652, and 0.02153338–0.02956331, respectively, for Cluster 2. The 10-day return forecasts showed relatively stable patterns. The Value at Risk (VaR) estimates at 90%, 95%, and 99% confidence levels indicate that, at the 99% confidence level, the highest potential loss was recorded for EXCL at −0.107764 in Cluster 1 and TOWR at −0.098700 in Cluster 2. Overall, Cluster 1 exhibited greater variation in risk levels than the relatively more homogeneous Cluster 2. The findings of this study are expected to provide a reference for investment risk analysis and serve as a consideration for investors in making investment decisions and managing ESG-based stock portfolios.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	ARIMA–GARCH, Clustering, Dynamic Time Warping, IDX ESG Leaders, K-Medoids, Value at Risk (VaR), Volatilitas Return Saham, ARIMA–GARCH, Clustering, Dynamic Time Warping, IDX ESG Leaders, K-Medoids, Stock Return Volatility, Value at Risk (VaR).
Subjects:	Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) Q Science > QA Mathematics > QA278.55 Cluster analysis Q Science > QA Mathematics > QA280 Box-Jenkins forecasting
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Actuaria > 94203-(S1) Undergraduate Thesis
Depositing User:	Yesinta Yesinta
Date Deposited:	17 Jul 2026 08:11
Last Modified:	18 Jul 2026 04:17
URI:	http://repository.its.ac.id/id/eprint/135354

Actions (login required)

View Item