FORECASTING WITH MISSPECIFIED MODELS

Abstract

This paper deals with the issue of forecasting with misspecified models. The models involved are the fixed-coefficient ARIMA model and the Random Coefficient Autoregressive model, with the random walk process as the model for its time-varying coefficients (RCA-RW model). The paper explores the gains in forecast accuracy by misspecifying time-varying coefficient processes with the parsimonious fixed-coefficient ARIMA model. In addition, the effect on the forecast performance of misspecifying the RCA-RW model to fixed-coefficient ARIMA process is also examined. In a simulation study, the forecast improvement for a first order Autoregresssive model with coefficients following a random walk process (the RCA-RW model) is discussed. From the study, it is found that the RCA-RW model provides superior forecasts over the AR(l) model only when there is sufficiently large variation in the coefficients over time. The time-varying coefficient methodology is used to analyze the Singapore and Thailand Consumer Price Indexes (CPI), and the forecast performance of this methodology is compared to that of the fixed-coefficient AR(I) model. The results obtained are found to be consistent with the simulation study.