Box-Cox Transformation for Exponential Smoothing With Application

Abstract

This article introduces a novel algorithm for incorporating power transformation into the estimation process of a Holt-Winters Seasonal model. The algorithm outlines a series of steps aimed at selecting the most appropriate power parameter estimate. This selection is achieved using the conventional Maximum Likelihood Estimation method in combination with various criteria for enhancing statistical modeling efficiency. Supplementary decision rules include assessing Mean Square Error, Mean Absolute Error, and conducting a p-value test for the normality of errors. The algorithm's effectiveness is demonstrated through its application to real-world data. Ultimately, the article affirms the feasibility of obtaining viable solutions for selecting the optimal power parameter.