Trianingtyas, Sakinah (2026) Perbandingan Model LSTM, BiLSTM, dan Stacked LSTM Dalam Peramalan Harga Saham Dan Estimasi Risiko Menggunakan CVaR Berbasis POT. Other thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 5006221057-Undergraduate_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
5006221057-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (5MB) | Request a copy |
Abstract
Investasi saham merupakan salah satu instrumen keuangan yang memiliki potensi Return tinggi, namun juga mengandung risiko akibat pergerakan harga yang fluktuatif. Oleh karena itu, diperlukan metode peramalan harga saham yang akurat serta pengukuran risiko ekstrem untuk mendukung pengambilan keputusan investasi. Penelitian ini bertujuan untuk memprediksi harga saham sektor non-cyclical yang tergabung dalam indeks IDX30, yaitu PT Indofood CBP Sukses Makmur Tbk (ICBP), PT Indofood Sukses Makmur Tbk (INDF), dan PT Unilever Indonesia Tbk (UNVR), serta mengestimasi risiko menggunakan Conditional Value at Risk (CVaR) berbasis Extreme Value Theory (EVT). Data yang digunakan berupa harga penutupan dan Return harian saham periode 2 Januari 2023 hingga 2 Januari 2026. Metode peramalan yang digunakan adalah Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (BiLSTM), dan Stacked Long Short-Term Memory (Stacked LSTM), dengan evaluasi model menggunakan Mean Absolute Percentage Error (MAPE). Hasil penelitian menunjukkan bahwa model terbaik untuk saham ICBP adalah BiLSTM dengan MAPE sebesar 1,2111%, sedangkan saham INDF dan UNVR memperoleh model terbaik menggunakan LSTM dengan MAPE masing-masing sebesar 1,2090% dan 2,2179%. Hasil peramalan 20 hari menunjukkan bahwa ICBP dan INDF cenderung mengalami tren kenaikan, sedangkan UNVR menunjukkan tren penurunan. Estimasi risiko berbasis EVT menunjukkan bahwa UNVR memiliki risiko ekstrem tertinggi dengan nilai CVaR 99% sebesar 0,118049, sedangkan INDF memiliki risiko ekstrem terendah sebesar 0,061197. Hasil backtesting menunjukkan bahwa model risiko lebih sesuai pada tingkat kepercayaan 99%.
==================================================================================================================================
Stock investment is one of the financial instruments that offers high potential Returns, but it also contains risk due to fluctuating price movements. Therefore, an accurate stock price forecasting method and extreme risk measurement are needed to support investment decision-making. This study aims to forecast the stock prices of non-cyclical sector companies listed in the IDX30 index, namely PT Indofood CBP Sukses Makmur Tbk (ICBP), PT Indofood Sukses Makmur Tbk (INDF), and PT Unilever Indonesia Tbk (UNVR), as well as to estimate risk using Conditional Value at Risk (CVaR) based on Extreme Value Theory (EVT). The data used in this study consist of daily closing prices and stock Returns from January 2, 2023, to January 2, 2026. The forecasting methods applied are Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (BiLSTM), and Stacked Long Short-Term Memory (Stacked LSTM), with model performance evaluated using Mean Absolute Percentage Error (MAPE). The results show that the best model for ICBP is BiLSTM, with a MAPE value of 1.2111%, while the best models for INDF and UNVR are LSTM, with MAPE values of 1.2090% and 2.2179%, respectively. The 20-day forecasting results indicate that ICBP and INDF tend to experience an upward trend, while UNVR shows a downward trend. Risk estimation using EVT-GPD shows that UNVR has the highest extreme risk, with a 99% CVaR value of 0.118049, while INDF has the lowest extreme risk, with a 99% CVaR value of 0.061197. The backtesting results indicate that the risk model is more suitable at the 99% confidence level.
Actions (login required)
 |
View Item |