Comparative Analysis of Ridge Truncated Spline Regression and Elastic Net Regression for Prediction Inflation Dynamics Based on Digital Payment Transactions in Indonesia

Authors

DOI:

https://doi.org/10.19139/soic-2310-5070-4263

Keywords:

Ridge Regression, Truncated Spline Regression, Elastic Net, E-money Transaction, Inflation Prediction

Abstract

This study investigates the dynamics of macroeconomic inflation in Indonesia by leveraging high-frequencyelectronic money transaction data (2021–2026). Given the severe multicollinearity and statistically properties inherent indigital payment indicators, this research proposes a comparative forecasting framework between Ridge Truncated SplineRegression (RTSR) and Elastic Net Regression (ENR). A time-series rolling-window cross-validation protocol (90:10,80:20, and 70:30 splits) was employed to rigorously evaluate out-of-sample predictive accuracy. Contrary to the initialhypothesis favoring the non-parametric flexibility of splines, empirical results robustly demonstrate that the Elastic Netbaseline significantly outperforms the proposed RTSR model across all forecasting scenarios. In the 70:30 testing split, ENRachieved an RMSE of 0.540 and a MASE of 1.399, effectively suppressing the severe extrapolation overfitting exhibitedby RTSR (RMSE: 5.143, MASE: 16.865). The Diebold-Mariano test confirms this predictive superiority is statisticallyhighly significant (p < 0.001). We conclude that while RTSR efficiently maps in-sample non-linearities, the dual linearregularization (L1 and L2) of ENR provides the critical algorithmic stability required to generalize forecasting during volatilemacroeconomic regime shifts, making it a superior tool for central bank policy formulation.

Downloads

Published

2026-11-18

How to Cite

Maylita Hasyim, Ahmad, I. S., & Siregar, N. R. A. A. (2026). Comparative Analysis of Ridge Truncated Spline Regression and Elastic Net Regression for Prediction Inflation Dynamics Based on Digital Payment Transactions in Indonesia. Statistics, Optimization & Information Computing. https://doi.org/10.19139/soic-2310-5070-4263

Issue

Section

Research Articles

Categories

Most read articles by the same author(s)