Muhammad Zarlis
Universitas Bina Nusantara Jakarta

Published : 1 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 1 Documents
Search

A Machine Learning-Based Early Warning System for Predicting Rupiah Exchange-Rate Pressure in Indonesia Welly Ramantika Ramantika; Muhammad Zarlis
Inkubis : Jurnal Ekonomi dan Bisnis Vol. 8 No. 2 (2026): INKUBIS Jurnal Ekonomi Dan Bisnis
Publisher : Politeknik Siber Cerdika Internasional

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.59261/inkubis.v8i2.316

Abstract

Background: Predicting economic vulnerability within the Twin Deficit Hypothesis (TDH) framework remains a significant challenge for developing countries due to the frequency mismatch between slow-moving macroeconomic fundamentals and rapidly changing financial market indicators. Objective: This study aims to address this gap by developing a hybrid Early Warning System (EWS) that integrates multi-frequency data to predict the probability of a Twin Deficit crisis in Indonesia with a one-month forecasting horizon. Methods: The dataset comprises macroeconomic and market data covering the period from 2010 to 2026. A mixed-frequency data alignment approach is employed to integrate quarterly fiscal and macroeconomic indicators, monthly fiscal deficit data, and daily market indicators, including the USD/IDR exchange rate, government bond yields, and global commodity prices. Results: Historical fiscal data gaps (2010–2017) are addressed using ARIMA-based stochastic interpolation. The XGBoost model, optimized using Borderline-SMOTE, achieves an accuracy of 79.17%, a recall rate of 87.50%, and a PR-AUC value of 0.9146 in a 24-month out-of-sample evaluation, outperforming Logit and Probit benchmark models. SHAP analysis identifies global Brent crude oil prices and domestic bond market stress as the dominant factors influencing the occurrence of Twin Deficit crises in Indonesia. Conclusion: These findings demonstrate that integrating high-frequency market signals enhances the detection of emerging fundamental vulnerabilities, providing an operational tool for fiscal and monetary authorities to support preventive policy coordination. The proposed machine-learning-based EWS, which combines mixed-frequency data integration with explainable artificial intelligence (XAI), offers a theoretically grounded and practically applicable framework for macroeconomic crisis monitoring in developing economies. This approach enables forward-looking and coordinated policy responses by Bank Indonesia and the Ministry of Finance of the Republic of Indonesia before fiscal and external imbalances escalate into a full-scale crisis.