Model Hybrid ARIMAX-QR dan QRNN untuk Peramalan Inflow dan Outflow Uang Kartal di Bank Indonesia Provinsi NTT dan Nasional

Permasalahan yang sering ditemui dalam peramalan time series adalah linearitas, nonlinearitas dan heteroskedastisitas. Metode untuk menganalisis permasalahan tersebut diantaranya metode Autoregressive Integrated Moving Average (ARIMA), Neural Network (NN) dan Quantile Regression (QR). Terdapat banyak penelitian dan pengembangan mengenai metode tersebut, namun penelitian mengenai perbandingan kinerja dari hybrid metode tersebut dalam menyelesaikan kasus data real masih terbatas. Oleh karena itu, penelitian ini dilakukan perbandingan kinerja antara metode hybrid ARIMAX-QR dan quantile Regression Neural Network (QRNN). Kedua metode digunakan pada studi kasus peramalan inflow dan outflow uang kartal Bank Indonesia Provinsi Nusa Tenggara Timur dan Nasional. Metode terbaik ditentukan berdasarkan kriteria Root Mean Squared Error (RMSE) dan Median Absolute Error (MdAE) terkecil. Berdasarkan studi kasus, metode ARIMAX menghasilkan RMSE dan MDAE out- sample terkecil untuk inflow uang kartal BI provinsi NTT sebesar 59,861 dan 36,122, sedangkan untuk outflow uang kartal BI provinsi NTT dan inflow serta outflow tingkat Nasional metode QRNN menghasilkan RMSE out-sample terkecil sebesar 99,009; 11340,110 dan 14786,96. Hasil tersebut menunjukkan bahwa metode ARIMAX memberikan kinerja peramalan lebih baik untuk meramalkan data inflow uang kartal Provinsi NTT yang bersifat heterogen linear. Sedangkan metode hybrid QRNN memberikan kinerja peramalan lebih baik untuk meramalkan data outflow uang kartal Provinsi NTT, data inflow serta outflow tingkat Nasional yang bersifat heterogen nonlinear =================================================================== The main problem that be frequently arise in time series analysis are nonlinearity and heteroscedasticity. There are some methods that could be used for analyzing these problems are Autoregressive Integrated Moving Average (ARIMA), Neural Network (NN) dan Quantile Regression (QR). There are a lot of researches and development on these methods, but the research focusing on performance comparison between the two methods applied in real case is still limited. Therefore, this study focuses on comparison between hybrid ARIMAX-QR, Quantile Regression Neural Network (QRNN) and ARIMAX. All methods are employed for modeling and forecasting the currency inflow and outflow data in Bank Indonesia. The results show that, ARIMAX method produced the smallest RMSE and MDAE out-samples for currency inflow in NTT province about 59.861 and 36.122, respectively. While for the currency outflow in NTT province and the national rate inflow and outflow, QRNN method resulted in the smallest RMSE out-sample of 99.009, 11340.110 and 14786.96, respectively. These results indicate that the ARIMAX method provides better forecasting performance for predicting the homogeneous linear data in currency inflow at NTT province. The QRNN hybrid method provides more accurate performance to forecast both currency outflow that contain non-linear heterogeneous data in NTT province and currency inflow and outflow data in National