Deteksi Krisis Keuangan Indonesia Menggunakan Model Hybrid Markov Switching Exponential Generalized Autoregressive Conditional Heteroskedasticity Dengan Efek Asimetris Dan Long Short-term Memory

Rochman, Alvian Noor (2026) Deteksi Krisis Keuangan Indonesia Menggunakan Model Hybrid Markov Switching Exponential Generalized Autoregressive Conditional Heteroskedasticity Dengan Efek Asimetris Dan Long Short-term Memory. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5003221115_Undergraduate_Thesis.pdf Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Krisis keuangan merupakan fenomena ekonomi yang dapat mengganggu stabilitas sistem keuangan dan perekonomian nasional sehingga diperlukan suatu sistem peringatan dini yang mampu mengidentifikasi potensi krisis secara akurat. Penelitian ini bertujuan untuk mendeteksi krisis keuangan di Indonesia serta mengembangkan model Early Warning System (EWS) menggunakan pendekatan hybrid Markov Switching Exponential Generalized Autoregressive Conditional Heteroskedasticity (MS-EGARCH) dan Long Short-Term Memory (LSTM). Data yang digunakan berupa data bulanan kurs USD terhadap rupiah, cadangan devisa, inflasi, dan suku bunga selama periode Januari 1990 hingga Desember 2024. Selain itu, data target berupa label kondisi krisis dan nonkrisis ditentukan berdasarkan periode krisis keuangan historis yang pernah terjadi di Indonesia. Tahapan pemodelan diawali dengan ARIMAX untuk memodelkan rata-rata kurs USD terhadap rupiah, kemudian dilanjutkan dengan EGARCH dan MS-EGARCH untuk memodelkan volatilitas, menangkap efek asimetris, serta mengidentifikasi perubahan rezim antara kondisi krisis dan nonkrisis. Hasil pemodelan menunjukkan bahwa model MS-EGARCH mampu mengidentifikasi periode krisis yang konsisten dengan kejadian krisis historis di Indonesia berdasarkan nilai smoothed probability. Selanjutnya, informasi yang dihasilkan dari model ARIMAX, EGARCH, dan MS-EGARCH digunakan sebagai fitur masukan pada model LSTM untuk mengklasifikasikan kondisi krisis dan nonkrisis. Hasil evaluasi pada data testing menunjukkan nilai accuracy sebesar 78,3%, precision sebesar 76,9%, recall sebesar 88,2%, dan F1-score sebesar 82,2%, yang menunjukkan bahwa model memiliki kemampuan yang baik dalam mengklasifikasikan kondisi krisis dan nonkrisis. Berdasarkan hasil peramalan untuk periode Januari hingga Desember 2025, kondisi nonkrisis diprediksi terjadi pada Januari hingga Maret 2025, sedangkan kondisi krisis diprediksi terjadi pada April hingga Desember 2025. Hasil penelitian ini menunjukkan bahwa pendekatan hybrid MS-EGARCH–LSTM berpotensi menjadi kerangka Early Warning System yang efektif untuk mendukung deteksi dini dan antisipasi krisis keuangan di Indonesia.
======================================================================================================================================
Financial crises are economic events that can threaten the stability of both the financial system and the national economy, highlighting the importance of developing an effective Early Warning System (EWS) to identify potential crises at an early stage. This study aims to detect financial crises in Indonesia and develop an EWS using a hybrid Markov Switching Exponential Generalized Autoregressive Conditional Heteroskedasticity (MS-EGARCH) and Long Short-Term Memory (LSTM) approach. Monthly data on the USD/IDR exchange rate, foreign exchange reserves, inflation, and interest rates from January 1990 to December 2024 were employed. Crisis and non-crisis target labels were constructed based on historical financial crisis periods in Indonesia. The modeling process begins with the ARIMAX model to capture the conditional mean of the exchange rate, followed by the EGARCH and MS-EGARCH models to model volatility, asymmetric effects, and regime transitions between crisis and non-crisis conditions. The MS-EGARCH model successfully identifies crisis periods that are consistent with historical financial crisis episodes in Indonesia through the estimated smoothed probabilities. Subsequently, the outputs generated by the ARIMAX, EGARCH, and MS-EGARCH models are utilized as input features for the LSTM model to classify crisis and non-crisis conditions. The evaluation on the testing dataset achieved an accuracy of 78.3%, precision of 76.9%, recall of 88.2%, and an F1-score of 82.2%, indicating that the proposed model demonstrates good classification performance in identifying financial crisis conditions. Furthermore, the forecasting results for the period from January to December 2025 predict non-crisis conditions from January to March 2025, followed by crisis conditions from April to December 2025. These findings suggest that the proposed hybrid MS-EGARCH–LSTM model has the potential to serve as an effective Early Warning System for supporting financial crisis monitoring and anticipation in Indonesia.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Krisis Keuangan, MS-EGARCH, LSTM, Makroekonomi, EWS, Financial Crisis, MS-EGARCH, LSTM, Macroeconomics, EWS
Subjects:	H Social Sciences > HA Statistics > HA30.3 Time-series analysis H Social Sciences > HB Economic Theory > Economic forecasting--Mathematical models. Q Science > Q Science (General) > Q325.5 Machine learning. Support vector machines. Q Science > QA Mathematics > QA274.7 Markov processes--Mathematical models.
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49201-(S1) Undergraduate Thesis
Depositing User:	Alvian Noor Rochman
Date Deposited:	31 Jul 2026 01:33
Last Modified:	31 Jul 2026 01:33
URI:	http://repository.its.ac.id/id/eprint/140064

Actions (login required)

View Item