This study aims to model and forecast inflation in Indonesia using a nonlinear regression approach, specifically the Logistic Smooth Transition Autoregressive (LSTAR) model. Inflation often exhibits nonlinear behavior due to different economic regimes, such as stable and crisis periods, which linear models fail to capture. Using monthly year-on-year inflation data from January 2009 to December 2022, this research begins with stationarity testing, nonlinearity identification using the Terasvirta test, LSTAR model parameter estimation, and diagnostic evaluation. The results confirm significant nonlinearity in Indonesian inflation. The best model is LSTAR(2) with the transition variable at a two-month lag (INF_{t-2}). This model effectively captures a smooth transition between two regimes: a low-inflation regime and a high-inflation regime, with an estimated threshold of 4.85%. In the low-inflation regime, inflation shows high persistence. In contrast, the high-inflation regime exhibits a mean-reverting characteristic, indicating a self-correcting mechanism or a policy response. Based on the Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) criteria, the LSTAR model demonstrates superior forecasting accuracy compared to the conventional linear ARIMA model. This study concludes that the LSTAR model is an effective and robust tool for modeling and forecasting inflation in Indonesia. The findings imply that Bank Indonesia should be more vigilant when inflation approaches the 4.85% threshold, as the economic dynamics may shift, requiring a different monetary policy response.
