Abstract
Inflation is marked by a decline in the domestic currency’s value and an increase in its exchange rate relative to foreign currencies. In Nigeria, this depreciation of the Naira has occurred alongside periods of rising inflation. Nonlinear time series models are particularly effective in capturing the complex dynamics of financial data, such as inflation rates. This study models Nigeria’s monthly inflation rate using three nonlinear approaches—Logistic Smooth Transition Autoregressive (LSTAR), Self-Excited Threshold Autoregressive (SETAR), and Artificial Neural Networks Time Series (NNETTs)—based on data from the Central Bank of Nigeria (CBN), covering the period from January 2005 to August 2023. Nonlinearity tests by Keenan and Tsay reveal that inflation rates between January 2016 and February 2024 follow a threshold nonlinear process, rejecting the null hypothesis of linearity and confirming the presence of structural breaks in the data. Visual inspection of the series further supports this. Among the models, the LSTAR model demonstrates superior performance with the lowest Akaike Information Criterion (AIC), Mean Absolute Percentage Error (MAPE), and Mean Square Error (MSE), making it the most effective for modeling the inflation rate. The LSTAR model identifies a critical threshold at 16.46, indicating a regime change in inflation behavior. Forecasts for September 2023 place the inflation rate at 25.42—well above the threshold—signaling that the economy has entered a higher-inflation regime. This trend continues through January 2024. The study concludes that the LSTAR model is a valuable tool for understanding regime-dependent inflation dynamics and recommends its adoption by analysts and policymakers for more accurate forecasting and strategic economic planning.