Forecasting of Time Series with Additive Periodical Terms by Exponential Smoothing

Abstract

The objective of this paper is to improve the transient response of the exponential smoothing with additive trend and periodical terms. A new exponential smoothing is derived from exponentially weighted least squares error criterion. Smoothing constants and variances of forecast errors are calculated numerically in the case that a trend is linear and a period is twelve. This method offers an advantage of having no tenency to oscillate for long periods with overshoot and undershoot in transient responses.