Abstract:
Inflation is the industrious and non stop ascent in the overall prices of any given commodity
in an economy. It is among the most macroeconomic variable described nonlinear behavior.
The aim of this study was also to model and forecast inflation in Ethiopia using nonlinear
models and to establish the existence of nonlinear patterns in the consumer price index. The
study utilized the secondary data collected from Monthly data of consumer price index for
inflation rate from January 1994 to December 2020 which was obtained from central
statistical Agency. The average monthly inflation rate was obtained 53.2800 with standard
deviation of 45.980 during the study period. The result showed that monthly rate of inflation
was characterized by a none constant mean and an unstable variance implying a non stationary the series and achieved stationarity by differecing orders. Nonlinearity tests result
based on Tsay tests showed non linearity of consumer price index and the SETAR (2,4,4) had
the minimum value for both Akakie Information Criteria and Bayesian Information criterion
among the camdidate models considered for the study. After modeling the inflation series was
made, the comparison of the forecast performance between the nonlinear time series models
and linear ARMA models based on forecast measure of mean absolute error (MAE), means
absolute percentage error (MAPE), and mean absolute scale error. This forcasting
performance comparison result showed that the nonlinear TAR family models suggest that the
nonlinear SETAR model outperform the linear ARMA models. The in-sample forecast using
the best-fit asymmetric model, that is SETAR (2,4,4) model indicates that the consumer price
index the series exhibits an upward trand year 2001 to 2010 and then almost similar livel
2011 to 2018 and then decrease at the end of study period. Based on this result it can be
recommended that, by using the Threshold Auto regressive Models policy makers would be
able to properly capture the price volatility persistence and hence forecasts and estimates
would be more accurate.
