Peramalan Harga Saham BBRI Menggunakan Metode Hybrid ARIMA-SVR

Abstract. PT Bank Rakyat Indonesia (BBRI) is one of the most popular investment instruments, but it has high stock price volatility due to global and domestic economic factors. This study aims to develop a BBRI stock price forecasting model using the Hybrid ARIMA–Support Vector Regression (ARIMA–SVR) approach, which combines the ability of ARIMA to capture linear patterns and SVR to capture non-linear patterns. The data used consists of daily closing prices of BBRI shares from June 17, 2020, to June 13, 2025, obtained from Yahoo Finance, with a total of 1,199 observations. ARIMA(1,1,0) and ARIMA(0,1,1) models were first constructed, and their residuals were processed using an SVR model with a Radial Basis Function kernel. The results of hyperparameter tuning indicate that the optimal SVR model is obtained with the combination of gamma (γ) = 0.5; cost (C) = 600; and epsilon (ε) of 0.0009 for ARIMA(1,1,0) and 0.0003 for ARIMA(0,1,1). Performance evaluation of the model shows that the Hybrid ARIMA(1,1,0)–SVR model provides the most accurate prediction results with a MAPE of 1.287% on the training data and 1.928% on the testing data. This study demonstrates that the Hybrid ARIMA–SVR approach is effective in handling complex and volatile stock time series data and can serve as a reference for more accurate and data-driven investment decision-making. Abstrak. PT Bank Rakyat Indonesia (BBRI) menjadi salah satu instrumen investasi yang banyak diminati, namun memiliki volatilitas harga saham yang tinggi akibat faktor ekonomi global dan domestik. Penelitian ini bertujuan membangun model peramalan harga saham BBRI menggunakan pendekatan Hybrid ARIMA–Support Vector Regression (ARIMA–SVR) yang menggabungkan kemampuan ARIMA dalam menangkap pola linier dan SVR dalam menangkap pola nonlinier. Data yang digunakan berupa harga penutupan harian saham BBRI dari 17 Juni 2020 hingga 13 Juni 2025 yang diperoleh dari Yahoo Finance, dengan total 1.199 observasi. Model ARIMA(1,1,0) dan ARIMA(0,1,1) dibangun terlebih dahulu dan hasil residunya diolah menggunakan model SVR dengan kernel Radial Basis Function. Hasil tuning hyperparameter, model SVR optimal diperoleh dengan kombinasi gamma (γ) = 0,5; cost (C) = 600; dan epsilon (ε) masing-masing sebesar 0,0009 untuk ARIMA(1,1,0) dan 0,0003 untuk ARIMA(0,1,1). Evaluasi kinerja model menunjukkan bahwa model Hybrid ARIMA(1,1,0)–SVR memberikan hasil prediksi paling akurat dengan MAPE sebesar 1,287% pada data training dan 1,928% pada data testing. Penelitian ini menunjukkan bahwa pendekatan Hybrid ARIMA–SVR efektif dalam menangani data deret waktu saham yang kompleks dan fluktuatif, serta dapat menjadi acuan dalam pengambilan keputusan investasi yang lebih akurat dan berbasis data.