Evaluasi Kinerja Prediksi Nilai Tukar Rupiah terhadap Yen Menggunakan Modifikasi Visibility Graph

Abstract

Penelitian ini memodelkan dan memprediksi pergerakan kurs jual rupiah terhadap yen menggunakan pendekatan visibility graph berbasis kemiripan. Metode visibility graph memetakan data time series menjadi graf tak berarah, dimana keterhubungan antar titik ditentukan berdasarkan visibilitas antar titik. Penelitian ini menggunakan data harian kurs IDR/JPY periode 1 Januari hingga 30 April 2024. Proses analisis dilakukan dengan membentuk visibility graph, menghitung matriks peluang transisi dan nilai kemiripan menggunakan local random walk, serta melakukan prediksi dengan tiga pendekatan: initial forecasting, amandatory forecasting dan modified amandatory forecasting. Evaluasi keakuratan dilakukan dengan indikator MAPE. Hasil penelitian menunjukkan bahwa amandatory forecasting memberikan hasil paling akurat dengan MAPE sebesar 0.792% yang lebih rendah dibanding kedua metode lainnya (0.939% dan 0.917%). Dengan demikian pendekatan visibility graph berbasis kemiripan terbukti efektif dalam memprediksi pergerakan kurs.

This study models and predict the movement of the rupiah selling exchange rate against the japanese yen using a similarity-based visibility graph approach. The visibility graph method transforms time series data into an undirected graph, where connections between nodes are determined based on their visibility. The study utilizes daily IDR/JPY exchange rate data from Januari 1 to April 30,2024. The analitycal process includes constructing the visibility graph, calculating the transition probability matrix and similarity values using local random walk, and performing predictions using three methods: initial forecasting, amandatory forecasting, and modified amandatory forecasting. The accuracy of each method is evaluated using the Mean Abosulte Percentage Error (MAPE). The result show that amandatory forecasting yields the most accurate prediction with a MAPE of 0.792%) which is lower than the other two methods (0.939% and 0.917%). Therefore, the similarity-based visibility graph approach proves to be effective in forecasting exchange rate movements.