Comparative Forecasting of Indonesian Stock Prices Using ARIMA and Support Vector Regression: A Statistical Learning Approach

Abstract

This study aims to compare the forecasting performance of Autoregressive Integrated Moving Average (ARIMA) and Support Vector Regression (SVR) models in predicting the monthly stock prices of PT. Telkom Indonesia (TLKM), one of Indonesia's leading state-owned enterprises. Utilizing historical data from January 2018 to December 2022, the models are evaluated based on their forecasting precision using Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE) as investor-grade performance metrics. The ARIMA (2,1,2) model was selected through Box–Jenkins methodology, while the SVR was optimized using a linear kernel and tuned hyperparameters. The results demonstrate that ARIMA outperforms SVR in both MAPE (2.01%) and RMSE (64.38), indicating better adaptability to the structured linear patterns found in the stock price series. Assumption-based forecasting simulations, extended to the year 2026, further suggest ARIMA’s relative stability, although future projections remain subject to structural and market uncertainties. The findings emphasize the continued relevance of classical statistical models in investment-grade forecasting, where investor-grade is defined as forecasting precision achieving MAPE values below 5%, supporting decision-support applications in emerging markets.