Forecasting Regional Economic Growth Using TVARX: Model Accuracy Evaluation in Banten Province

Abstract

Forecasting regional economic performance is essential for supporting timely and responsive policy planning. This study aims to forecast the Gross Regional Domestic Product at constant prices (GRDP) in Banten Province for the second to fourth quarters of 2025 using the Time-Varying Autoregressive model with Exogenous Variables (TVARX). The model incorporates household final consumption expenditure, gross fixed capital formation, exchange rates, and export values as exogenous variables. Model performance was evaluated by comparing combinations of training-testing data proportions (90:10, 80:20, 70:30, and 60:40) and two estimation approaches (local constant and local linear), using the Mean Absolute Percentage Error (MAPE) as the predictive accuracy metric. All variables were transformed into logarithmic form and differenced to ensure stationarity. The results indicate that the model using a 90:10 data split and the local linear estimation approach yielded the most accurate prediction, with the lowest MAPE value of 0.6%. The best-performing model was then applied to forecast out-of-sample GRDP CP for the next three quarters, with its year-on-year growth subsequently analyzed. These findings are expected to serve as a basis for data-driven economic analysis and support macroeconomic planning that is responsive to short-term structural dynamics.