Abstract
An estimation of the smoothing parameters is formulated in which the data of demand shows the periodicity. The minimization of squared forecasting errors is adopted as a criterion of the estimation. Three algorithms for the minimization are attained through the use of i) Gradient method, ii) Gauss-Newton method, and iii) CRST method. In the latter case, several constraints on the smoothing parameters can be taken into account. Each algorithm is able to become an element of a forecasting system with the pre-estimation procedure and the test of forecasting errors.
