Pengembangan Singular Spectrum Analysis-Vector Dengan Pendekatan State-Dependent Model Studi Kasus: Peramalan Ekspor Indonesia

Sasmita, Yoga (2024) Pengembangan Singular Spectrum Analysis-Vector Dengan Pendekatan State-Dependent Model Studi Kasus: Peramalan Ekspor Indonesia. Doctoral thesis, Institut Teknologi Sepuluh Nopember.

Text 7003201005-Dissertation.pdf - Accepted Version Restricted to Repository staff only until 1 October 2026. Download (5MB) | Request a copy

Abstract

Data deret waktu finansial seperti nilai ekspor seringkali memiliki karakteristik pola nonlinier seperti adanya unsur structural break, tren, harmonik dan random serta fluktuasi antarwaktu yang tinggi. Salah satu metode analisis yang dapat mengakomodir permasalahan tersebut adalah Singular Spectrum Analysis-Vector (SSAV). SSA memiliki kemampuan yang baik dalam mendekomposisi data menurut pola unsurnya (tren, osilasi dan noise). Namun pada SSAV standar belum dapat menangani data yang memiliki structural break dengan baik, oleh karena itu penelitian ini mengembangkan metode SSAV yang menggunakan pendekatan median-based dan State-Dependent Model (SDM). Dalam implementasi SDM, penaksiran vektor state yang digunakan adalah Extended Kalman Filter (EKF) yang disingkat dengan mbSSAV-SDM EKF dan Ensemble Kalman Filter (EnKF) yang disingkat dengan mbSSAV-SDM EnKF. Untuk menentukan nilai awal ( , dan ) yang digunakan dalam SDM penelitian ini menggunakan teknik Bootstrap SSA. Dalam menentukan kombinasi parameter optimum penelitian ini mengimplementasikan beberapa kombinasi rentang parameter pada metode yang dikembangkan kemudian dipilih dari RMSE yang terkecil. Evaluasi akurasi diukur dengan RMSE dan SMAPE yang dihitung dengan prosedur Cross Validation. Kajian simulasi menunjukkan metode mbSSAV-SDM EKF memiliki akurasi paling baik dibandingkan dengan metode mbSSAV-SDM EnKF, SSAR-SDM EKF, SETAR dan ETS. Kajian terapan studi ini menggunakan data bulanan nilai total ekspor Indonesia dari 1993 hingga 2022. Evaluasi akurasi pada kajian terapan menunjukkan metode mbSSAV-SDM EKF memiliki akurasi paling baik dibandingkan mbSSAV-SDM EnKF, SSAR-SDM EKF, SETAR dan ETS. Dengan RMSE metode mbSSAV-SDM EKF pada titik prediksi data testing h = 1 hingga h = 12 berturut-turut sebesar 0,0964, 0,1193, 0,1240, 0,1232, 0,1680, 0,1576, 0,1879, 0,1273, 0,1522, 0,2225, 0,2409 dan 0,2567. SMAPE pada titik peramalan h = 1 hingga h = 12 berturut-turut sebesar 0,6850, 0,9831, 0,9725, 1,0004, 1,4460, 1,3231, 1,4769, 1,0895, 1,2837, 2,0125, 1,9499 dan 2,2088. =====================================================================================================================================
Financial time series data, such as export values, often have nonlinear pattern characteristics, such as structural breaks, trends, harmonics, and noise elements, as well as high inter-temporal fluctuations. One analytical method that can address these problems is the Singular Spectrum Analysis-Vector (SSAV). SSA has a good ability to decompose data according to its elemental patterns (trends, oscillations and noise). However, the standard SSAV cannot handle data with structural breaks well. Therefore, this study develops SSAV methods that use median-based and State-Dependent Model (SDM) approaches. In the implementation of SDM, the state vector estimators used are the extended Kalman filter (EKF), which is abbreviated as mbSSAV-SDM EKF, and the ensemble Kalman filter (EnKF) abbreviated as mbSSAV-SDM EnKF. To determine the initial values ( , dan ) used in SDM, this study employs the bootstrapping SSA technique. Furthermore, to determine the optimum parameter combination, this study implemented several combinations of parameter ranges in the developed method and then selected the smallest RMSE. Accuracy evaluation was performed using the RMSE and SMAPE values calculated via cross validation procedure. The simulation studies show that the mbSSAV-SDM EKF method has the best accuracy compared with the mbSSAV-SDM EnKF, SSAR-SDM EKF, SETAR, and ETS methods. The applied study uses monthly data on Indonesia’s total export value from 1993 to 2022. The accuracy evaluation in the applied study shows that the mbSSAV-SDM EKF method exhibits the best accuracy compared with the mbSSAV-SDM EnKF, SSAR-SDM EKF, SETAR, and ETS. With the RMSE of the mbSSAV-SDM EKF method at the point of prediction of testing data h = 1 to h = 12 was 0,0964, 0,1193, 0,1240, 0,1232, 0,1680, 0,1576, 0,1879, 0,1273, 0,1522, 0,2225, 0,2409, and 0,2567, respectively. SMAPE at forecasting points h = 1 to h = 12 was 0,6850, 0,9831, 0,9725, 1,0004, 1,4460, 1,3231, 1,4769, 1,0895, 1,2837, 2,0125, 1,9499, and 2,2088, respectively.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	median-based Singular Spectrum Analysis Vector, structural break, State-Dependent Model, Ekspor Indonesia. median-based, Singular Spectrum Analysis Vector, structural break, State-Dependent Model, Indonesian Export.
Subjects:	H Social Sciences > HA Statistics > HA30.3 Time-series analysis
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Statistics > 49001-(S3) PhD Thesis
Depositing User:	Yoga Sasmita
Date Deposited:	10 Aug 2024 13:31
Last Modified:	13 Aug 2024 02:02
URI:	http://repository.its.ac.id/id/eprint/115297

Actions (login required)

View Item