Estimasi Risiko Carbon Emission Futures Menggunakan Long Short-Term Memory berbasis Non-Overlapping Block Bootstrap dan Moving Block Bootstrap

'Ilma, Shofiyah Muqodimatul (2026) Estimasi Risiko Carbon Emission Futures Menggunakan Long Short-Term Memory berbasis Non-Overlapping Block Bootstrap dan Moving Block Bootstrap. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5006221019-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (8MB) | Request a copy

Abstract

Pada tahun 2024 total emisi dunia tercatat meningkat sebesar 1,3% dibandingkan tahun sebelumnya. Oleh karena itu, pengendalian emisi gas rumah kaca menjadi semakin penting. Salah satu mekanisme yang mendukung upaya tersebut adalah pasar karbon, termasuk perdagangan kontrak carbon emission futures pada European Union Emissions Trading System (EU ETS). Namun, pergerakan harga kontrak carbon emission futures bersifat volatil sehingga menimbulkan risiko yang perlu diestimasi secara akurat. Penelitian ini bertujuan membandingkan kinerja metode Non-Overlapping Block Bootstrap (NBB) dan Moving Block Bootstrap (MBB) dalam estimasi risiko carbon emission futures menggunakan model Long Short-Term Memory (LSTM) sebagai dasar pembentukan estimasi Value at Risk (VaR) yang divalidasi menggunakan uji Kupiec Proportion of Failures (POF). Data yang digunakan berupa harga penutupan harian carbon emission futures EU ETS periode 4 Januari 2021 hingga 31 Desember 2025. Setelah dilakukan pemodelan LSTM dengan berbagai kombinasi hyperparameter, model LSTM terbaik diperoleh pada kombinasi window length 20, hidden neuron 16, dan dropout rate 0,3 yang menghasilkan nilai MAPE training sebesar 1,8352% dan MAPE testing sebesar 0,8284%. Berdasarkan model tersebut, estimasi VaR dilakukan menggunakan metode NBB dan MBB. Hasil estimasi menunjukkan bahwa metode NBB menghasilkan 27 pelanggaran pada tingkat kepercayaan 95% dan 99%, sedangkan metode MBB menghasilkan 27 pelanggaran pada tingkat kepercayaan 95% dan 24 pelanggaran pada tingkat kepercayaan 99%. Hasil backtesting menunjukkan bahwa nilai VaR kedua metode tidak valid. Meskipun demikian, metode MBB menunjukkan kinerja yang lebih baik dengan nilai likelihood ratio sebesar 136,261445, lebih rendah dibandingkan metode NBB sebesar 161,204409. Evaluasi lebih lanjut melalui beberapa skenario perbaikan model menunjukkan bahwa penerapan transformasi Box-Cox pada metode MBB mampu meningkatkan kinerja estimasi risiko dibandingkan skenario lainnya. Dengan demikian, metode MBB dengan transformasi Box-Cox memberikan hasil terbaik dalam estimasi risiko carbon emission futures berbasis LSTM.
=====================================================================================================================================
In 2024, global carbon emissions increased by 1.3% compared to the previous year. Therefore, efforts to control greenhouse gas emissions have become increasingly important. One mechanism supporting these efforts is the carbon market, including trading carbon emission futures contracts under the European Union Emissions Trading System (EU ETS). However, carbon emission futures prices are highly volatile, creating risks that require accurate estimation. This study aims to compare the performance of the Non-Overlapping Block Bootstrap (NBB) and Moving Block Bootstrap (MBB) methods in estimating carbon emission futures risk using a Long Short-Term Memory (LSTM) model as the basis for Value at Risk (VaR) estimation, validated through the Kupiec Proportion of Failures (POF) test. The data consists of daily closing prices of EU ETS carbon emission futures from 4th January 2021 to 31st December 2025. After developing LSTM models with various hyperparameter combinations, the best performing model was obtained with a window length of 20, 16 hidden neurons, and a dropout rate of 0.3, resulting in training and testing MAPE values of 1.8352% and 0.8284%. Based on this model, VaR estimation was performed using the NBB and MBB methods. The results indicate that the NBB method produced 27 violations at both the 95% and 99% confidence levels, while the MBB method produced 27 violations at the 95% confidence level and 24 violations at the 99% confidence level. Backtesting results showed that VaR estimations from both methods were not valid. Nevertheless, the MBB method demonstrated better performance, with a likelihood ratio value of 136.261445, lower than the NBB value of 161.204409. Further evaluation through several model improvement scenarios revealed that applying the Box-Cox transformation to the MBB method improved risk estimation performance. Therefore, the MBB method with Box-Cox transformation provided the best results for LSTM based carbon emission futures risk estimation.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Carbon Emission Futures, Long Short-Term Memory, Moving Block Bootstrap, Non-Overlapping Block Bootstrap, Value at Risk
Subjects:	H Social Sciences > HA Statistics > HA30.3 Time-series analysis H Social Sciences > HG Finance > HG4915 Stocks--Prices 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:	Shofiyah Muqodimatul 'ilma
Date Deposited:	20 Jul 2026 02:01
Last Modified:	20 Jul 2026 02:01
URI:	http://repository.its.ac.id/id/eprint/135466

Actions (login required)

View Item