A Multiple-Regime Econometric Model of Inflation for Mauritius

Abstract

This paper estimates a semi-multivariate dynamic model of Mauritian inflation, using monthly data over the period January 1976 - December 2001, which captures the significant nonlinearity and asymmetry present in the inflation process. Starting from a linear autoregressive distributed lag (ARDL) model, asymmetric nonlinearity is found and modelled using a logistic smooth transition regression (LSTR) model. The LSTR model, while producing an improvement in fit over the ARDL model, contains significant autocorrelation and fails tests of parameter constancy and remaining nonlinearity. A logistic multiple-regime smooth transition regression (LMRSTR) model is then estimated, constituting a marked improvement over the LSTR model and hence the linear ARDL model. A simulation exercise produces interesting findings on the dynamic effects of the Rs./$ exchange rate on Mauritian inflation and demonstrates the superiority of the LMRSTR model over the conventional ARDL model.