SIGNAL EXTRACTION TO FORECAST TIME SERIES FROM REPEATED SAMPLE SURVEYS

摘要

This study proposes an approach for forecasting time series obtained from repeated sample surveys. The approach applies a signal extraction technique as a preprocessing step to remove noises, due to sampling errors, from the surveyed time series data. Box-Jenkins ARIMA is used to forecast using the preprocessed signal. The intended result is to show that this approach performs better in terms of RMSE. An extensive computational simulation study is designed to support our hypotheses on various types of time series. Partial results show that for time series data from repeated sample surveys, forecasting should be done using signals extracted from signal extraction approach in order to reduce forecasting errors.