Regularized Vector Autoregressive Model for the Assessment of Macroeconomic Variables Associated with Inflation in Nigeria

Abstract

This study investigates the determinants and forecasting performance of inflation in Nigeria using Regularized Vector Autoregressive (RVAR) models. While standard VAR models are widely applied in macroeconomic analysis, they are often affected by conditional multicollinearity and overparameterization, particularly in small samples, leading to unstable estimates and poor forecasting accuracy. To address these limitations, the study employs Ridge VAR, LASSO VAR, and Elastic Net VAR, and compares their performance with that of the standard VAR framework. Monthly data on inflation, currency in circulation, savings rate, and crude oil prices were analyzed. Unit root tests showed that most variables are non-stationary in levels and are modeled in first differences. Empirical results indicate that oil price shocks have a strong but short-lived impact on inflation, while the savings rate and currency circulation explain a substantial share of inflation variability. Forecast evaluation shows that LASSO and Elastic Net VAR models outperform the standard VAR, reducing forecast RMSE by approximately 43% and achieving RMSSE values of 0.651 compared to 1.137 for the standard VAR. The findings demonstrate that regularized VAR models provide a more reliable framework for inflation forecasting and policy analysis in Nigeria.