Forecasting economic time series with measurement error

Abstract

Many variables used in economic forecasting are recorded with measurement error (ME). It is therefore found that an autoregressive model without exclusion of ME from observed time series may fail to correctly detect any periodicity contained and this results in poor forecasting performances. The purpose of this paper is to propose a model-selection method for forecasting economic time series with ME. In this method the existence or nonexistence of ME is determined by evaluating the values of the Akaike information criterion (AIC) of a battery of alternative models with and without ME. The results of forecasting 26 business cycle indicators in Japan are shown in order to demonstrate the efficacy of the proposed method.