Perbandingan Gaussian Process Regression, Backpropagation Neural Network, dan Long Short-Term Memory untuk Prediksi Saham dalam Optimisasi Portofolio

Arsanti, Bening Nazhifa (2026) Perbandingan Gaussian Process Regression, Backpropagation Neural Network, dan Long Short-Term Memory untuk Prediksi Saham dalam Optimisasi Portofolio. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5006221088-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Investasi merupakan aktivitas penting dalam perekonomian modern, di mana penempatan dana pada instrumen keuangan bertujuan memperoleh return optimal dengan risiko terkendali. Saham perbankan BUMN, khususnya PT Bank Mandiri Tbk (BMRI), PT Bank Negara Indonesia Tbk (BBNI), dan PT Bank Rakyat Indonesia Tbk (BBRI), menjadi fokus penelitian ini karena memiliki kapitalisasi pasar besar, likuiditas tinggi, serta berperan penting dalam menjaga stabilitas pasar modal Indonesia. Namun, volatilitas harga saham yang dipengaruhi faktor ekonomi, politik, maupun sentimen pasar menjadikan prediksi harga saham sebuah tantangan. Penelitian ini bertujuan mengembangkan model prediksi harga saham berbasis machine learning serta mengintegrasikannya dalam pembentukan portofolio optimal dan estimasi risiko investasi. Tiga algoritma digunakan, yaitu Gaussian Process Regression (GPR), Backpropagation Neural Network (BPNN), dan Long Short-Term Memory (LSTM). Data penelitian berupa harga penutupan harian saham periode Agustus 2022 hingga Juli 2025. Evaluasi model menggunakan metrik Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), dan Mean Absolute Error (MAE). Hasil penelitian menunjukkan bahwa BPNN memberikan performa terbaik dengan MAPE sebesar 1,298% untuk BMRI, 1,335% untuk BBNI, dan 1,440% untuk BBRI. Setelah prediksi diperoleh, dilakukan optimisasi portofolio menggunakan Model Mean-Variance Optimization dengan tingkat risk aversion (γ) sebesar 5, menghasilkan komposisi optimal yaitu BMRI 42,46%, BBNI 36,73%, dan BBRI 20,81%. Estimasi risiko portofolio menggunakan Value at Risk (VaR) berbasis simulasi Monte Carlo dengan 1.000 replikasi pada tingkat kepercayaan 95% menghasilkan VaR portofolio sebesar 2,4%, lebih rendah dibandingkan VaR individual masing-masing saham, memvalidasi keberhasilan diversifikasi dalam menurunkan risiko investasi.
======================================================================================================================================
Investment is a crucial activity in the modern economy, where fund allocation in financial instruments aims to obtain optimal returns with controlled risk. State-owned bank stocks, particularly PT Bank Mandiri Tbk (BMRI), PT Bank Negara Indonesia Tbk (BBNI), and PT Bank Rakyat Indonesia Tbk (BBRI), are the focus of this research due to their large market capitalization, high liquidity, and important role in maintaining the stability of Indonesia's capital market. However, stock price volatility influenced by economic, political, and market sentiment factors makes stock price prediction a challenge. This research aims to develop machine learning-based stock price prediction models and integrate them into optimal portfolio formation and investment risk estimation. Three algorithms are employed: Gaussian Process Regression (GPR), Backpropagation Neural Network (BPNN), and Long Short-Term Memory (LSTM). The research data consists of daily closing stock prices from August 2022 to July 2025. Model evaluation uses Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and Mean Absolute Error (MAE) metrics. The research results show that BPNN provides the best performance with MAPE of 1.298% for BMRI, 1.335% for BBNI, and 1.440% for BBRI. After predictions are obtained, portfolio optimization is performed using the Mean-Variance Optimization Model with a risk aversion level (γ) of 5, resulting in optimal composition of BMRI 42.46%, BBNI 36.73%, and BBRI 20.81%. Portfolio risk estimation using Value at Risk (VaR) based on Monte Carlo simulation with 1,000 replications at 95% confidence level yields a portfolio VaR of 2.4%, lower than the individual VaR of each stock, validating the success of diversification in reducing investment risk.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Backpropagation Neural Network, Gaussian Process Regression, Long Short-Term Memory, Optimisasi Portofolio, Value at Risk. Backpropagation Neural Network, Gaussian Process Regression, Long Short-Term Memory, Portofolio Optimization, Value at Risk.
Subjects:	Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) Q Science > QA Mathematics > QA353.K47 Kernel functions (analysis) Q Science > QA Mathematics > QA76.87 Neural networks (Computer Science)
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Actuaria > 94203-(S1) Undergraduate Thesis
Depositing User:	Bening Nazhifa Arsanti
Date Deposited:	12 Jan 2026 05:33
Last Modified:	12 Jan 2026 05:33
URI:	http://repository.its.ac.id/id/eprint/129487

Actions (login required)

View Item