A Comparison of Short-Term Forecasting Models

摘要

Seven short-term forecasting models, two using least-squares estimation methods and five employing variations of the exponentially weighted moving average method, are compared in their relative ability to produce minimum error variance forecasts for seven simulated time series. Each series was generated to enable one of the forecast models to be the least squared error predictor. A comparison methodology is developed which facilitates forecast model performance through the measurement of model specification errors. A computer program is presented which may be modified to accept real time series and which permits the forecast models to be ranked in order of their relative specification error. (AUTHOR)