A long memory model with mixed normal GARCH for US inflation data

Abstract

We introduce a time series model that captures both long

memory and conditional heteroskedasticity and assess its ability to

describe the US inflation data. Specifically, the model allows for

long memory in the conditional mean formulation and uses a

normal mixture GARCH process to characterize conditional

heteroskedasticity. We find that the proposed model yields a good

description of the salient features, including skewness and

heteroskedasticity, of the US inflation data. Further, the performance

of the proposed model compares quite favorably with, for example,

ARMA and ARFIMA models with GARCH errors characterized by

normal, symmetric and skewed Student-t distributions.