Analisis Perbandingan Peramalan Return Saham Menggunakan Pendekatan Hybrid GARCH-family Dan Deep Feed-Forward Neural Network

Raharjo, Rachel Ayuningtyas Putri (2025) Analisis Perbandingan Peramalan Return Saham Menggunakan Pendekatan Hybrid GARCH-family Dan Deep Feed-Forward Neural Network. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5006211011-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy

Abstract

Volatilitas return saham yang tinggi menjadi tantangan dalam aktivitas investasi, terutama di pasar modal Indonesia yang terdiri atas papan utama, pengembangan, dan akselerasi. Setiap papan mencerminkan karakteristik perusahaan yang berbeda, termasuk risiko dan stabilitas pergerakan harga saham. Oleh karena itu, metode peramalan yang akurat sangat diperlukan guna mendukung pengambilan keputusan investasi. Penelitian ini bertujuan membandingkan performa model Deep Feed-Forward Neural Network (DFFNN) dengan pendekatan hybrid yang menggabungkan model DFFNN dan GARCH-family, yang mencakup GARCH, EGARCH, dan APGARCH. Studi dilakukan menggunakan data harian harga penutupan saham dari BBRI, BNLI, dan WGSH yang mewakili masing-masing papan selama periode Januari 2022 hingga Februari 2025. Evaluasi dilakukan dengan menggunakan Akaike’s Information Criterion (AIC) untuk menilai akurasi model GARCH dan Mean Squared Error (MSE) untuk menilai akurasi model DFFNN. Hasil penelitian menunjukkan bahwa saham ketiga saham dapat dimodelkan dengan GARCH dan EGARCH yang mengindikasikan adanya volatility clustering. Pada pemodelan APGARCH, hanya saham WGSH yang memiliki parameter signifikan secara keseluruhan. Berdasarkan hasil pelatihan model yang dilakukan menunjukkan bahwa model DFFNN memiliki MSE yang paling rendah untuk saham BBRI dan BNLI sebesar 0,0000059450 dan 0,0000013806 dengan kombinasi hidden neuron (50, 10). Hal berbeda ditunjukkan oleh saham WGSH yang memiliki nilai MSE terkecil pada pendekatan hybrid EGARCH-DFFNN sebesar 0,000923876 dengan kombinasi hidden neuron (10, 10). Model terbaik digunakan untuk peramalan selama 10 periode mendatang yang dapat menjadi dasar pertimbangan investor dalam merumuskan strategi investasi.
===================================================================================================================================
High return volatility poses a significant challenge in investment activities, particularly in the Indonesian capital market, which is segmented into the main board, development board, and acceleration board. Each board reflects distinct characteristics of the listed companies, including differences in risk profiles and the stability of stock price movements. Therefore, accurate forecasting methods are crucial to support effective investment decision-making. This study aims to compare the performance of the Deep Feed-Forward Neural Network (DFFNN) model with a hybrid approach that combines DFFNN with GARCH-family models, including GARCH, EGARCH, and APGARCH. The analysis was conducted using daily closing price data from BBRI, BNLI, and WGSH—each representing a different board—over the period from January 2022 to February 2025. Model accuracy was evaluated using Akaike’s Information Criterion (AIC) for GARCH models and Mean Squared Error (MSE) for DFFNN models. The results indicate that all three stocks can be modeled using GARCH and EGARCH, suggesting the presence of volatility clustering. In the APGARCH model, only the WGSH stock exhibited statistically significant parameters overall. Model training results show that DFFNN achieved the lowest MSE for BBRI and BNLI, with values of 0.0000059450 and 0.0000013806, respectively, using a hidden neuron configuration of (50, 10). In contrast, the WGSH stock yielded the lowest MSE in the hybrid EGARCH-DFFNN approach, with a value of 0.000923876 using a hidden neuron configuration of (10, 10). The best-performing models were then used to forecast stock returns for the next 10 periods, providing a potential basis for investors in formulating investment strategies.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Deep Feed-Forward Neural Network, Generalized Autoregressive Conditional Heteroskedasticity, Return Saham, Volatilitas, Deep Feed-Forward Neural Network, Generalized Autoregressive Conditional Heteroskedasticity, Stock Return, Volatility
Subjects:	H Social Sciences > HA Statistics > HA30.3 Time-series analysis H Social Sciences > HG Finance > HG4529 Investment 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:	Rachel Ayuningtyas Putri Raharjo
Date Deposited:	28 Jul 2025 02:25
Last Modified:	28 Jul 2025 02:25
URI:	http://repository.its.ac.id/id/eprint/121890

Actions (login required)

View Item