Penerapan Moving Block Bootstrap Pada Peramalan Nilai Tukar Rupiah Terhadap USD Menggunakan Neural Basis Expansion Analysis For Time Series

Wijaya, Verliana Nathaline (2026) Penerapan Moving Block Bootstrap Pada Peramalan Nilai Tukar Rupiah Terhadap USD Menggunakan Neural Basis Expansion Analysis For Time Series. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5006221009-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Nilai tukar Rupiah terhadap USD merupakan salah satu indikator ekonomi yang bersifat fluktuatif dan menimbulkan risiko nilai tukar, khususnya bagi perusahaan asuransi yang memiliki eksposur terhadap aset dan kewajiban dalam mata uang asing. Fluktuasi nilai tukar yang tinggi dapat meningkatkan ketidakpastian dalam pengambilan keputusan investasi dan pengelolaan risiko keuangan. Hal ini mendorong perlunya metode peramalan yang akurat serta estimasi risiko yang mampu menggambarkan potensi kerugian akibat perubahan nilai tukar. Penelitian ini bertujuan untuk meramalkan nilai tukar menggunakan Neural Basis Expansion Analysis for Time Series (N-BEATS) dan mengestimasi risiko nilai tukar melalui Value at Risk (VaR) dengan pendekatan Moving Block Bootstrap (MBB). Data yang digunakan berupa data historis harian nilai tukar Rupiah terhadap USD periode 2 Januari 2015 hingga 27 Februari 2026 yang diperoleh dari Bank Indonesia. Peramalan nilai tukar dilakukan menggunakan N-BEATS, sedangkan estimasi risiko dilakukan dengan membangkitkan 50 data bootstrap menggunakan MBB, yang selanjutnya diramalkan kembali menggunakan N-BEATS. Hasil peramalan tersebut digunakan dalam perhitungan VaR pada tingkat kepercayaan 95% dan 99%. Kinerja model dievaluasi menggunakan Mean Absolute Percentage Error (MAPE), sedangkan validitas estimasi risiko dinilai melalui proses backtesting. Hasil penelitian menunjukkan bahwa kombinasi hyperparameter terbaik terdiri atas lookback period 40, stacks 2, blocks 3, block-layers 3, width 32, dan batch size 32, dengan hyperparameter width dan batch size yang paling berpengaruh. Model tersebut menghasilkan nilai MAPE training sebesar 0,6545% dan MAPE testing sebesar 0,1758%, yang menunjukkan kemampuan prediksi sangat baik. Hasil peramalan untuk 10 hari kerja berikutnya menunjukkan nilai tukar diperkirakan bergerak relatif stabil pada sekitar Rp16.777 - Rp16.912 per USD. Estimasi VaR menghasilkan model yang valid berdasarkan hasil backtesting, dimana nilai Likelihood Ratio pada tingkat kepercayaan 95% dan 99% lebih kecil dibandingkan nilai kritis Chi-Square. Hasil tersebut menunjukkan bahwa N-BEATS dan MBB mampu menghasilkan peramalan yang akurat serta estimasi risiko yang valid.
===============================================================================================================================
The Rupiah/USD exchange rate is an economic indicator that fluctuates over time and creates exchange rate risk, particularly for insurance companies with exposure to assets and liabilities denominated in foreign currencies. High volatility increases uncertainty in investment decisions and financial risk management. Therefore, accurate forecasting and risk estimation methods are needed to measure potential losses caused by exchange rate movements. This study aims to forecast the Rupiah/USD exchange rate using Neural Basis Expansion Analysis for Time Series (N-BEATS) and estimate risk using Value at Risk (VaR) with the Moving Block Bootstrap (MBB) approach. The data consist of daily historical exchange rates from January 2, 2015, to February 27, 2026, obtained from Bank Indonesia. Forecasting was performed using N-BEATS, while risk estimation was conducted by generating 50 bootstrap datasets through MBB and forecasting them using the same model. The forecasts were then used to calculate VaR at the 95% and 99% confidence levels. Model performance was evaluated using Mean Absolute Percentage Error (MAPE), while risk estimation validity was assessed through backtesting. The results indicate that the best hyperparameter combination consists of a lookback period of 40, 2 stacks, 3 blocks, 3 block-layers, a width of 32, and a batch size of 32, with width and batch size being the most influential hyperparameters. The model achieved a training MAPE of 0.6545% and a testing MAPE of 0.1758%, indicating excellent predictive performance. The forecast for the next 10 business days suggests a relatively stable exchange rate ranging from IDR 16,780 to IDR 16,920 per USD. Backtesting results confirmed the validity of the VaR model, as the Likelihood Ratio values at both confidence levels were lower than the critical Chi-Square value. These findings show that N-BEATS and MBB provide accurate forecasts and valid risk estimates.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Moving Block Bootstrap, N-BEATS, Nilai Tukar Rupiah, Peramalan, Value at Risk (VaR), Exchange Rate, Forecasting, Moving Block Bootstrap, N-BEATS, Value at Risk (VaR)
Subjects:	Q Science > QA Mathematics Q Science > QA Mathematics > QA276 Mathematical statistics. Time-series analysis. Failure time data analysis. Survival analysis (Biometry) 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:	Verliana Nathaline Wijaya
Date Deposited:	17 Jul 2026 08:00
Last Modified:	17 Jul 2026 08:00
URI:	http://repository.its.ac.id/id/eprint/135371

Actions (login required)

View Item