Multi-Horizon Forecasting Harga Beras melalui Horse Race Model Linear Teregularisasi di Jawa Bagian Tengah dan Timur

Abstract

Volatilitas harga beras di Jawa Tengah, Jawa Timur, dan Daerah Istimewa Yogyakarta memicu efek rambatan spasial yang mendistorsi ekuilibrium pasar. Penelitian ini mengevaluasi model machine learning teregularisasi, yaitu AR, ARDL, dan FA-ARDL, untuk meramalkan harga enam kualitas beras. Menggunakan data 137 pasar nasional yang difokuskan pada 28 pasar dengan horizon peramalan 1, 5, 10, dan 22 hari ke depan, hasil penelitian menunjukkan bahwa model AR berpenalti LASSO unggul pada horizon 1 hari dengan kemampuan menyaring noise lokal, sedangkan model FA-ARDL Nasional berpenalti Elastic Net lebih stabil pada horizon 10 sampai 22 hari karena mampu menangkap spillover spasial. Kualitas beras Low dan Medium didominasi oleh model FA-ARDL Nasional, sementara kualitas Super didominasi oleh model AR. Temuan ini mendukung pengembangan sistem peringatan dini untuk operasi pasar, penguatan Kerja Sama Antardaerah (KAD), serta fleksibilitas anggaran melalui Belanja Tidak Terduga (BTT) minimal 4,8%.

Rice price volatility in Central Java, East Java, and the Special Region of Yogyakarta generates spatial spillover effects that disrupt market equilibrium. This study evaluates regularized machine learning models, namely AR, ARDL, and FA-ARDL, to forecast prices across six rice quality categories. Using data from 137 national markets, focused on 28 markets, and forecasting horizons of 1, 5, 10, and 22 days ahead, the results show that the AR model penalized by LASSO performs best at the 1-day horizon by filtering local noise, while the National FA-ARDL model penalized by Elastic Net provides more stable forecasts at horizons of 10 and 22 days by capturing spatial spillovers. Low and medium grades of rice are primarily forecasted by the National FA-ARDL model, whereas super-grade rice is dominated by the AR model. These findings support the development of an early warning system for market operations, interregional cooperation (KAD), and budget flexibility through a minimum 4.8% contingency allocation under Unexpected Expenditure (BTT) funds.