Empirical Study of Inflation Modelling Using Threshold Autoregressive Models in Nigeria

Nigeria's inflation dynamics display asymmetric behaviour across high and low inflation regimes, suggesting that linear autoregressive models are inadequate for capturing the full complexity of price level evolution in the country's inflationary environment. This study applies threshold autoregressive models, including the self-exciting threshold autoregressive model and the band-threshold autoregressive model, to monthly CPI inflation data from the National Bureau of Statistics spanning January 2003 to December 2023. Threshold estimation is performed using the concentrated least squares approach of Hansen, with confidence intervals constructed through bootstrap resampling. Tests for threshold nonlinearity reject the null of linearity at the 1 percent significance level, with the estimated inflation threshold at 16.8 percent separating the low and high regimes. In the low-inflation regime, inflation exhibits stronger mean reversion with a half-life of 4.2 months, while in the high-inflation regime, mean reversion is weaker with a half-life of 9.7 months, indicating greater persistence of price pressures once inflation breaks through the threshold. Out-of-sample forecast evaluation over a 24-month holdout sample shows that the threshold autoregressive model outperforms linear ARIMA and GARCH-mean models with 14 percent lower mean squared forecast error. The asymmetric dynamics identified have direct implications for monetary policy calibration at the Central Bank of Nigeria. Keywords: threshold autoregressive model, inflation dynamics, Nigeria, nonlinearity, monetary policy.