Hakiki, Ferigo Taufani Tri (2026) Pengaruh Berita Perubahan Iklim terhadap Volatilitas Bersama Sektor Saham Minyak dan Gas di Indonesia menggunakan Hybrid Ensemble ARX-GARCH Type-LSTM dengan Autoencoder. Masters thesis, Institut Teknologi Sepuluh Nopember.
![[thumbnail of 6003241034-Master_Thesis.pdf]](http://repository.its.ac.id/style/images/fileicons/text.png) |
Text
6003241034-Master_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (7MB) | Request a copy |
Abstract
Penelitian ini bertujuan untuk membentuk volatilitas bersama (common volatility) saham sektor minyak dan gas di Indonesia (PGAS, MEDC, AKRA, ENRG) serta menganalisis pengaruh berita iklim global terhadapnya. Model yang digunakan adalah Hybrid Ensemble ARX-GARCH Family-LSTM dengan Autoencoder. Data harian periode 2010-2025 diproses melalui tahapan transformasi data kebentuk log-return, pengendalian pengaruh faktor eksternal menggunakan model ARX, estimasi volatilitas individual dengan GARCH, EGARCH, dan GJRGARCH, dilanjutkan kepemodelan Hybrid dengan LSTM, pembentukan Ensemble of errors melalui meta learning neural networks, ekstraksi common volatility menggunakan Autoencoder, dan estimasi model ARX (3) dengan variabel agregat serta klaster berita iklim MCCC. Hasil penelitian menunjukkan bahwa model Hybrid Ensemble skenario ALL THREE to AVG memberikan performa terbaik dengan R² tertinggi serta RMSE dan MAE terendah pada estimasi volatilitas untuk masing-masing saham. Sedangkan dalam pembentukan volatilitas Bersama Autoencoder mampu menjelaskan 86,7% varians keempat saham. Berita iklim global secara langsung tidak berpengaruh signifikan terhadap common volatility, namun interaksi MCCC × WTI berpengaruh signifikan negatif. Dari keempat tema berita iklim (Business, Environmental, Societal, Research), tema Research merupakan yang paling dominan melalui interaksinya dengan guncangan WTI dan secara marginal dengan IHSG. Kesimpulannya, berita iklim mempengaruhi volatilitas sektor migas secara tidak langsung melalui interaksi dengan guncangan pasar. Model Hybrid Ensemble mampu meningkatkan akurasi prediksi volatilitas, dan Autoencoder efektif mereduksi multiple volatilitas menjadi satu representasi bersama. Implikasinya, manajemen risiko sektor energi perlu mempertimbangkan efek moderasi berita iklim terhadap guncangan harga minyak serta menggunakan model Hybrid Ensemble untuk prediksi volatilitas yang lebih akurat.
=============================================================================================================================================
This study aims to construct the common volatility of Indonesia’s oil and gas (O&G) sector stocks (PGAS, MEDC, AKRA, and ENRG) and to analyze the influence of global climate news on such volatility. The proposed approach employs a Hybrid Ensemble ARX–GARCH Family–LSTM model integrated with an Autoencoder. Daily data spanning the 2010–2025 period were processed through several stages, including data transformation into log returns, controlling for external factor effects using the ARX model, individual volatility estimation using GARCH, EGARCH, and GJR-GARCH models, Hybrid modeling with LSTM, construction of an Ensemble of errors through meta-learning neural networks, extraction of common volatility using an Autoencoder, and estimation of an ARX(3) model incorporating both aggregate and clustered Media and Climate Change Concerns (MCCC) climate news variables. The results indicate that the Hybrid Ensemble model under the ALL THREE to AVG scenario achieved the best performance, yielding the highest R² values and the lowest RMSE and MAE values across the volatility estimates of all four stocks. The Autoencoder explained 86.7% of the variance across the four stocks, with the largest contribution originating from MEDC (39.0%). Global climate news does not exert a statistically significant direct effect on common volatility; however, the interaction between MCCC and WTI Shocks exhibits a significant negative effect. Among the four climate news themes (Business, Environmental, Societal, and Research), the Research theme emerged as the most dominant, primarily through its interaction with WTI Shocks and marginally with IHSG Shocks. In conclusion, climate news influences the volatility of Indonesia’s O&G sector indirectly through its interaction with market Shocks. The Hybrid Ensemble model improves volatility forecasting accuracy, while the Autoencoder effectively reduces multiple individual volatilities into a single common representation. These findings imply that energy sector risk management should consider the moderating effect of climate news on oil price Shocks and adopt Hybrid Ensemble models for more accurate volatility forecasting
Actions (login required)
 |
View Item |