Least Squares Spline Estimation Method in Semiparametric Time Series Regression for Predicting Indonesia Composite Index

Abstract

The Least Squares Spline (LS-Spline) method offers a flexible approach for modeling fluctuating time series data by adaptively positioning knots at points of structural change. This study develops an LS-Spline estimation method for the Semiparametric Time Series Regression (STSR) model, combining an autoregressive structure as the parametric component and multiple nonparametric functions to capture nonlinear effects. The model is applied to predict the Indonesia Composite Index (ICI), a key indicator of sustainable economic growth. In this framework, the ICI at lag-1 is modeled parametrically, while the BI Rate and Inflation are modeled nonparametrically. Four data splitting schemes 6, 12, 18, and 24 months of testing data are used to evaluate forecasting performance over short, medium, and long term horizons. Results show that the LS-Spline STSR model consistently achieves high predictive accuracy, with MAPE and sMAPE below 10% and MASE below 1. Residual diagnostics using ACF and PACF confirm that the model satisfies the white noise assumption. These findings emphasize the potential of the LS-Spline STSR model as an economic forecasting tool that can support policies related to one of poin Sustainable Development Goals (SDGs), namely sustainable economic growth.