Analisis Penggunaan Additional Regressor Pada Model Prophet untuk Peramalan Nilai Tukar Rupiah Terhadap Dolar Amerika Serikat

Abstract

Fluktuasi nilai tukar Rupiah terhadap Dolar Amerika Serikat dipengaruhi oleh berbagai faktor ekonomi sehingga diperlukan model peramalan yang mampu mengakomodasi informasi dari peubah eksternal. Penelitian ini bertujuan menganalisis penggunaan U.S. Dollar Index (DXY) dan Indeks Harga Saham Gabungan (IHSG) sebagai additional regressor pada model Prophet untuk peramalan nilai tukar USD/IDR serta mengevaluasi pengaruhnya terhadap kinerja model. Data yang digunakan berupa data harian nilai tukar USD/IDR, DXY, dan IHSG selama Januari 2015–Desember 2025. Tahapan penelitian meliputi identifikasi lag effect menggunakan korelasi Pearson, pemodelan Prophet dalam empat skenario, serta optimasi hyperparameter menggunakan Tree-Structured Parzen Estimator (TPE) yang dievaluasi melalui time series cross-validation dengan skema expanding window. Sebagai analisis pendukung, digunakan model Vector Autoregressive (VAR) melalui uji Granger Causality, Impulse Response Function (IRF), dan Forecast Error Variance Decomposition (FEVD). Hasil penelitian menunjukkan bahwa model Prophet baseline menghasilkan kinerja terbaik dengan MAPE 1,23%, MAE 201,60, dan RMSE 259,74, sedangkan penambahan DXY dan IHSG belum meningkatkan akurasi peramalan. Analisis VAR menunjukkan DXY dan IHSG memiliki hubungan dinamis yang signifikan dengan nilai tukar, meskipun sebagian besar variasi nilai tukar masih dipengaruhi oleh pergerakan historisnya. Peramalan menggunakan model Prophet baseline menunjukkan nilai tukar USD/IDR diperkirakan bergerak relatif stabil selama sepuluh hari kerja berikutnya.

The fluctuations in the Indonesian Rupiah exchange rate against the U.S. Dollar are influenced by various economic factors, highlighting the need for forecasting models capable of incorporating information from external variables. This study aims to analyze the use of the U.S. Dollar Index (DXY) and the Jakarta Composite Index (JCI) as additional regressors in the Prophet model for forecasting the USD/IDR exchange rate and to evaluate their impact on forecasting performance. The study used daily data on the USD/IDR exchange rate, DXY, and JCI from January 2015 to December 2025. The research procedure included lag effect identification using Pearson correlation, Prophet modeling under four different scenarios, and hyperparameter optimization using the Tree-Structured Parzen Estimator (TPE), with model performance evaluated through expanding-window time series cross-validation. As a supporting analysis, a Vector Autoregressive (VAR) model was employed to examine the dynamic relationships among variables using Granger causality, Impulse Response Function (IRF), and Forecast Error Variance Decomposition (FEVD). The results indicate that the baseline Prophet model achieved the best forecasting performance, with a MAPE of 1.23%, MAE of 201.60, and RMSE of 259.74, while the inclusion of DXY and JCI did not improve forecasting accuracy. The VAR analysis revealed significant dynamic relationships between DXY, JCI, and the exchange rate, although most of the exchange rate variation was explained by its own historical movements. Forecasts generated by the baseline Prophet model suggest that the USD/IDR exchange rate will remain relatively stable over the next ten business days.