Hierarchical Forecasting Of Currency Inflow And Outflow In Bank Indonesia Based On Hybrid Arimax-Ann Model

Prayoga, I Gede Surya Adi (2016) Hierarchical Forecasting Of Currency Inflow And Outflow In Bank Indonesia Based On Hybrid Arimax-Ann Model. Masters thesis, Institut Teknologi Sepuluh Nopember Surabaya.

[thumbnail of 1314201031-Master_Thesis.pdf]
Preview
Text
1314201031-Master_Thesis.pdf - Published Version

Download (4MB) | Preview

Abstract

Tujuan penelitian ini adalah untuk mendapatkan metode terbaik untuk meramalkan inflow dan outflow uang kartal bulanan di Bank Indonesia. Data inflow maupun outflow merupakan data deret waktu hirarki dengan beberapa tingkatan, yaitu data tingkat kantor perwakilan wilayah, kepulauan, dan tingkat nasional. Untuk meramalkan data hirarki diperlukan perlakuan khusus agar dihasilkan ramalan yang mengikuti struktur hirarki. Penelitian ini membandingkan performa dari beberapa metode hirarki, antara lain bottom-up, top-down dan kombinasi optimal. Setiap metode tersebut diterapkan pada ramalan dasar yang diperoleh menggunakan model terbaik antara ARIMAX dan hybrid ARIMAX-ANN. Performa masing-masing metode dibandingkan berdasarkan kriterie RMSE untuk data out-of-sample. Hasil yang diperoleh menunjukkan bahwa dalam mendapatkan ramalan dasar, metode hybrid dapat meningkatkan akurasi sebanyak 97,8 dan 87,0 persen dari total 46 series untuk masing-masing inflow dan outflow. Besar peningkatan akurasinya mencapai 10,26 dan 10,65 persen masing-masing untuk inflow dan outflow. Selain itu, banyaknya data yang mengandung heteroskedastisitas juga berkurang setelah diterapkan metode hybrid. Dalam peramalan hirarki, metode top-down dan kombinasi optimal baik digunakan jika ramalan dasar didapatkan hanya dari model ARIMAX. Jika digunakan model terbaik untuk masing-masing data, metode bottom-up adalah metode terbaik untuk data hirarki inflow maupun outflow. Selanjutnya, ramalan interval dihitung dengan menggunakan varians error yang diestimasi berdasarkan MSE dan GARCH, masing-masing untuk data dengan homoskedastisitas dan heteroskedastisitas. Ramalan interval berdasarkan GARCH cenderung lebih sempit atau lebih presisi dibandingkan dengan ramalan interval berdasarkan MSE, terutama untuk periode yang jauh ke depan
=============================================================================================
The aim of this study is to find the best method for forecasting monthly currency inflow and outflow data in Bank Indonesia. The inflow or outflow data is a hierarchical time series that has some levels, i.e. branch offices, islands, and national level. Hierarchical time series forecasting requires special treatment to make the forecasts follow the hierarchy structure. This study compared the performance of hierarchical forecasting methods including bottom-up, top-down and optimal combination method. Those methods were implemented on the base forecast obtained individually by using the best model between ARIMAX and hybrid ARIMAX-ANN. The method performance were evaluated based on out-of-sample RMSE. The results showed that in obtaining base forecasts of 46 series of each inflow and outflow, the hybrid method could increase the accuracy of 97.8 and 87.0 percent of ARIMAX model for inflow and outflow respectively. The accuracy improvement of ARIMAX models were up to 10.26 and 10.65 percent respectively on inflow and outflow series.Moreover, the hybrid method also reduced the number of series that has heteroscedasticity. In the hierarchical forecasting, the top-down and optimal combination method performed well if the base forecasts were only based on ARIMAX. If the base forecasts were obtained by using the best model for each series, which are more accurate, the bottom-up method could outperformed the other methods for both hierarchical currency inflow and outflow data. Finally, the interval of final forecast were constructed by using the error variance, which are estimated based on MSE and GARCH respectively for series with homoscedasticity and heteroscedasticity. The interval forecast based on GARCH model will not certainly become wider over time and tend to be narrower or more precise compared to the interval based on MSE, especially for long period of forecasting

Item Type: Thesis (Masters)
Additional Information: RTSt 515.535 Pra h
Uncontrolled Keywords: hierarchical time series, bottom-up, top-down, optimal combination, ARIMAX, hybrid ARIMAX-ANN, interval forecast
Subjects: Q Science > QA Mathematics > QA280 Box-Jenkins forecasting
Divisions: Faculty of Mathematics, Computation, and Data Science > Statistics > 49101-(S2) Master Thesis
Depositing User: EKO BUDI RAHARJO
Date Deposited: 16 Jul 2020 01:42
Last Modified: 29 Apr 2024 08:26
URI: http://repository.its.ac.id/id/eprint/76384

Actions (login required)

View Item View Item